Journal of Educational Psychology
1990, Vol. 82, No. 4,715-726
Copyright 1990 by the American Psychological Association, Inc.
0022-O663/90/SOO.75
When Is an Illustration Worth Ten Thousand Words?
Richard E. Mayer
Joan K. Gallini
University of California, Santa Barbara
Department of Educational Psychology,
University of South Carolina
In three experiments, students read expository passages concerning how scientific devices work,
which contained either no illustrations (control), static illustrations of the device with labels for
each part (parts), static illustrations of the device with labels for each major action (steps), or
dynamic illustrations showing the "off" and "on" states of the device along with labels for each
part and each major action (parts-and-steps). Results indicated that the parts-and-steps (but not
the other) illustrations consistently improved performance on recall of conceptual (but not
nonconceptual) information and creative problem solving (but not verbatim retention), and these
results were obtained mainly for the low prior-knowledge (rather than the high prior-knowledge)
students. The cognitive conditions for effective illustrations in scientific text include appropriate
text, tests, illustrations, and learners.
The two major media for communicating scientific information to students are words and pictures. In spite of the
traditional bias toward verbal over visual forms of instruction,
a growing research base suggests that text illustrations can
have important effects on student learning (Levie & Lentz,
1982; Mandl & Levin, 1989; Mayer, 1989a; WiUows &
Houghton, 1987). This article is based on the premise that
techniques for enhancing students' visual learning of scientific
information represent a relatively untapped potential for improving instruction.
When is an illustration likely to improve student learning?
What makes a good illustration? Which features of illustrations make them effective? How can we use illustrations to
improve students' learning from expository prose? These are
the kinds of questions that motivate our study. In this introduction, we examine the idea that one function of illustrations
(which we call explanative illustrations) is to help readers
build runnable mental models and that several conditions
must be met for explanative illustrations to be effective.
tions can help the reader remember key information in a text;
(4) organization—illustrations can help the reader organize
information into a coherent structure; and (5) interpretation—illustrations can help the reader understand the text. In
this article we focus exclusively on illustrations aimed at
serving the interpretation function, which we call explanative
illustrations.
For this study we set our instructional goal as the promotion
of learner understanding of how scientific systems work, as
measured by qualitative reasoning about scientific systems
(Bobrow, 1985). We define a "good illustration** as one that
promotes the reader's understanding of how a scientific system works. Therefore, we focus our study on explanative
illustrations, that is, illustrations that promote interpretation
processes.
Based on theories of mental models (de Kleer & Brown,
1985; Gentner & Stevens, 1983; Kieras & Bovair, 1984;
Larkin & Simon, 1987; White & Frederiksen, 1987), we
identify two major features of illustrations that could help
learners build runnable mental models: system topology and
component behavior. System topology refers to the portrayal
of each major component within the structure of the system.
For example, the parts illustrations shown in Figures 1 and 2
present the system topologies of a brake system and a pump
system, respectively. In Figure I, the labeled components
within the braking system include the tube, wheel cylinder,
small pistons, brake shoe, and brake drum.
Component behavior refers to the portrayal of each major
state that each component can be in and the relation between
a state change in one component and state changes in other
components. For example, the parts-and-steps illustrations in
Figures 1 and 2 show alternative states of the components in
a braking system and a pump system, respectively. In Figure
1, the parts-and-steps illustrations show the "before" and
"after" states of the tube, smaller pistons, brake shoes, and
wheel: the fluid in the tube can be compressed or not, small
pistons can be back or forward, the brake shoe can be pressing
against or away from the drum, and the wheel can be spinning
or still. In addition, the relations among state changes are
emphasized in the steps illustrations and the parts-and-steps
illustrations: if the fluid in the tube is compressed, the smaller
Cognitive Functions of Illustrations
What makes a good illustration? In order to define what
we mean by "good illustration" we must further specify "good
for what?" Levin (1981; Levin, Anglin, & Carney, 1987) has
proposed five functions of text illustrations: (1) decoration—
illustrations can help the reader enjoy the textbook by making
it more attractive (but without being relevant to the text); (2)
representation—illustrations can help the reader visualize a
particular event, person, place, or thing (such as found commonly in narrative passages); (3) transformation—illustraThe authors gratefully acknowledge the assistance of Jeffrey Lee
Steinberg in running subjects and Bill Lee in conducting statistical
analyses. The first author was mainly responsible for Experiments 1
and 2. The second author was mainly responsible for Experiment 3.
Correspondence concerning this article should be addressed to
Richard E. Mayer, Department of Psychology, University of California, Santa Barbara, California 93106, or to Joan Gallini, Department
of Educational Psychology, University of South Carolina, Columbia,
South Carolina 29208.
715
716
RICHARD E. MAYER AND JOAN K. GALL1NI
PARTS ILLUSTRATION
TUBE
WHEEL CYLINDER
SMALLER PISTON
STEPS ILLUSTRATION
When the driver steps on the car's brake pedal...
A piston moves forward inside the master cylinder (not shown)
The piston forces brake fluid out
of the master cylinder and through
the tubes to the wheel cylinder.-
BRAKE DRUM
In the wheel cylinder, the increase
in fluid pressure makes a set of
smaller pistons move.
BRAKE SHOE
When the brake shoes press against
the drum, both the drum and the
wheel stop or slow down.
PARTS AND STEPS ILLUSTRATION
(TEXT INCLUDES BOTH OF THE ABOVE ILLUSTRATIONS)
Figure 1. Illustrations of a brake system. (Adapted from The World Book Encyclopedia. Vol. 2, p.
463, 1987, Chicago: World Book, Inc. Copyright 1990 by World Book, Inc. by permission of the
publisher.)
pistons move forward; if the smaller pistons move forward,
the brake shoe will push against the drum; if the shoe presses
against the drum, the spinning wheel will stop.
Current theories of mental models suggest the potential
effectiveness of qualitative models in teaching students about
scientific systems (de Kleer & Brown, 1985; Kieras & Bovair,
1984; Gentner & Gentner, 1983; White & Frederiksen, 1987).
Such models are constructed to impart visual representations
of how a system looks and how it functions under various
state changes. A straightforward implication of this work is
that experience with qualitative models will allow students to
acquire alternative conceptualizations of how a system works
and to develop skill in predicting the causal behavior of the
system. Unfortunately, empirical research in this area is limited, but even more importantly, it generally ignores the
educationally relevant issues of how illustrations can be designed and used to promote the acquisition of runnable
mental models. Our line of research builds on existing theories
of mental models, but focuses on the instructional question
"When is an illustration most likely to be effective in promoting scientific understanding?"
Conditions for Effective Illustrations
As summarized in Figure 3, Mayer (1989b) has proposed
four conditions that must be met for illustrations to be effective in promoting understanding of scientific text: (1) exploitative text—the text must present a cause-and-effect system
that allows for qualitative reasoning (Bobrow, 1985); (2) sensitive tests—the performance measures must evaluate the
learners understanding and qualitative reasoning about the
system; (3) explanative illustrations—the illustrations must
help the learner build a runnable mental model of the system;
and (4) inexperienced learners—the students must not spon-
taneously engage in active learning processes such as the
construction of a runnable mental model of the system. In
summary, the text, tests, illustrations, and learners must all
be appropriate for the instructional goal of fostering meaningful learning.
In this article we report on three experiments that each test
six predictions generated from the model in Figure 3. In each
experiment, students read a scientific text that contains explanative illustrations, nonexplanative illustrations, or no illustrations. Then students write down what they remember
and answer questions about the text.
First, we examine two predictions' concerning the test
condition shown in Figure 3.
1. Explanative illustrations improve recall of explanative
information but not nonexplanative information. This prediction follows from schema theory (Rumelhart & Ortony, 1977):
conceptual understanding depends on formulating a schema
or knowledge structure about how a system operates. Specifically, the schema describes relations among parts of the
system including how a change in one part of the system
affects a change in another part of the system (i.e., explanative
information). Supplemental factual information (i.e., nonexplanative information) might temporarily be stored in memory as discrete pieces of knowledge but generally would not
be assimilated within the schema of the system.
2. Explanative illustrations improve creative problem solving but not verbatim retention. This prediction is based on the
idea that the most direct measure of conceptual understanding
is a problem solving test that requires what Bobrow (1985)
calls "qualitative reasoning about physical systems" (p. 1) or
' Predictions 1 through 4 are based on the idea that condition 4 is
met, namely that the learners are inexperienced; therefore, data
analysis concerning these predictions in each experiment was based
only on data from low prior-knowledge learners.
WHAT MAKES A GOOD ILLUSTRATION?
PARTS ILLUSTRATION
717
STEPS ILLUSTRATION
HANDLE
As the rod is pulled out,
air passes through the piston
and fills the areas between
the piston and outlet valve.
• PISTON
INLET VALVE
OUTLET VALVE
H06E
As the rod is pushed in,
the inlet valve closes
and the piston forces air
through the outlet valve.
PARTS AND STEPS ILLUSTRATION
t
\
•HANDLE
As the rod is pulled out,
- As the rod is pushed in
-air passes through the piston
PISTON
INLET VALVE
• the inlet valve closes
OUTLET VALVE
and fills the area between the
piston and the outlet valve.
and the piston forces air through
the outlet valve.
Figure 2. Illustrations of a pump system, (Adapted from The World Book Encyclopedia, Vol. 15, p.
794, 1987, Chicago: World Book, Inc. Copyright 1990 by World Book, Inc. by permission of the
publisher.)
system topology and component behavior are likely to foster
what de Kleer & Brown (1983) call "running a causal model"
the learner's building of a runnable mental model. Learners
(p. 158). In contrast, verbatim retention is irrelevant to the
who have built a runnable mental model are more likely to
goal of meaningful understanding.
recall information about how one state change affects another
Second, we examine two predictions concerning the illusbut less likely to remember irrelevant facts, as compared to
tration condition shown in Figure 3.
3. Explanative illustrations are more effective in improving students who have not built mental models.
4. Explanative illustrations are more effective in improving
conceptual recall than nonexplanative illustrations. This preproblem-solving performance than nonexplanative illustradiction aims to isolate the features of effective illustrations,
tions. Following the same line of reasoning as presented for
namely, only illustrations that explicitly concretize both the
718
RICHARD E. MAYER AND JOAN K. GALLINI
Experiment I (Brakes)
Explanative x/ text?
(nonexplanative text)
(nonconceptual test>
(nonexplanative illustrations)
(high prior-knowledge learners)
Figure 3. Four conditions for effective illustrations.
Prediction 3, learners who possess runnable mental models
should be able to engage in qualitative reasoning better than
learners who do not (White & Frederiksen, 1987).
The final two predictions are based on the learner condition
in Figure 3.
5. Explanative illustrations will improve recall of explanative information for low prior-knowledge learners but not for
high prior-knowledge learners. This prediction is based on the
idea that high prior-knowledge learners come to the learning
situation with a repertoire of techniques for strategically using
their domain knowledge of mechanical systems (Alexander &
Judy, 1988; McDaniel & Pressley, 1987), whereas low priorknowledge learners do not. High prior-knowledge learners
possess strategies for visually representing information in the
text and can coordinate imagery strategies with other procedures. These techniques help high prior-knowledge learners
focus on the explanative information during learning so special illustrations are not needed. In contrast, imagery strategy
research (McDaniel & Pressley, 1987) demonstrates that low
prior-knowledge learners, who lack skill in visualizing on their
own, will be more likely to profit from imagery-based instructional supplements than high prior-knowledge learners.
6. Explanative illustrations will improve problem-solving
performance for low prior-knowledge learners but not for high
prior-knowledge learners. Following Prediction 5, high priorknowledge learners spontaneously build runnable mental
models that can be used in problem solving so special illustrations are not needed.
Method
Subjects and design. The subjects were 96 college students recruited from the psychology subject pool at the University of California, Santa Barbara. Twenty-four students served in each of the four
treatment groups: (1) the no illustrations group read a booklet about
braking systems that contained no illustrations; (2) the parts illustrations group read a booklet that contained illustrations showing the
major parts within each type of braking system; (3) the steps illustrations group read a booklet containing illustrations showing the major
actions occurring for each type of braking system; and (4) the partsand-steps illustrations group received a booklet with both types of
illustrations. Half of the students in each group were low priorknowledge students who rated their knowledge of automobile mechanics as "very little" and reported that they never had performed
any automobile repairs, whereas the other half were high priorknowledge students who rated their knowledge of automobile mechanics as more than "very little" and reported having performed
minor car maintenance (such as changing oil or installing a spark
plug).
Materials. The materials consisted of four versions of a booklet
on braking systems, a subject questionnaire, and three posttests, each
typed on 8.5 x 11 inch sheets of paper.
The text in the booklet was taken from the World Book Encyclopedia's (1987) entry for "brakes," containing 750 words and 95 idea
units. The text included an explanation of how each of four braking
systems operate: mechanical braking (such as on bicycles), hydraulic
disk braking systems (such as on the front wheels of cars), hydraulic
drum braking systems (such as on the rear wheels of cars), and air
braking systems (such as on trains and trucks). For each type of
braking system, the text included a description of how a change in
the status of one part affects changes in the status of other parts (i.e.,
what Mayer, 1984, has called explanative information) as well as
factual information such as historical information or descriptions of
the materials used to make the parts (i.e., what Mayer, 1984, has
called nonexplanative information).
The no-illustrations version contained only the text; the partsillustrations version added illustrations showing the status of each of
the four braking systems before the brakes are applied, along with
labels for the major parts (using names repeated from the text); the
steps-illustrations version added illustrations showing the status of
each of the four braking systems after the brakes are applied, along
with a list of the major changes that occur (using wording repeated
from the text); the parts-and-steps illustrations version added both
the parts and the steps illustrations for each of the four braking
systems. Illustrations appeared on the same page and below their
corresponding paragraphs. A portion of the text is given in Table 1
and examples of each type of illustration are given in Figure 1.
The subject questionnaire solicited information concerning the
students' experiences with automobile mechanics and repair. For
example, students were asked to use a 5-point scale ranging from
"very little" to "very much" for the item: "Please put a check mark
indicating your knowledge of car mechanics and repair." In another
item, students were asked: "Please place a check mark next to the
things you have done." The list included having a drivers' license,
putting air into a tire, changing a tire, changing oil, installing spark
plugs, and replacing brake shoes. The recall posttest asked students
to "write down all you can remember from the passage you have just
read" and to "pretend that you are writing an encyclopedia for
beginners."
The problem-solving posttest consisted of five questions, each
printed on a separate sheet: (a) Why do brakes get hot? (b) What
could be done to make brakes more reliable, that is, to make sure
719
WHAT MAKES A GOOD ILLUSTRATION?
Table 1
Portion of Text on Brakes (With Explanative
Information in Italics)
Each student's recall protocol was broken down into idea
units; credit was given if an idea unit in the student's recall
protocol conveyed the same meaning or contained the same
key words as an idea unit in the passage (see Mayer, 1985,
Hydraulic brakes use various fluids instead of levers or cables. In
1989b). The verbatim retention test was scored by tallying the
automobiles, the brake fluid is in chambers called cylinders. Metal
number of correct answers. The problem-solving test was
tubes connect the master cylinder with wheel cylinders located near
scored by tallying the number of correct answers produced
the wheels. When the driver steps on the car's brake pedal, a piston
for each problem, as described by Mayer (1989b). Examples
moves forward inside the master cylinder. The piston forces brake
of acceptable answers for the five problem-solving questions
fluid out of the master cylinder and through the tubes to the wheel
cylinders. In the wheel cylinders, the increase in fluid pressure makes are: (a) Heat is caused by friction, pressure, rubbing, or
a set of smaller pistons move. These smaller pistons activate either
pressing a stationary object (such as a brake shoe) against a
drum brakes or disk brakes. Most automobiles have drum brakes on
rotating object (such as a rim or disk); (b) Reliability could
the rear wheels and disk brakes on the front wheels.
be
increased by adding a backup braking system, by using
Drum brakes consist of a cast-iron drum and a pair of semicircular
thicker shoes, tougher tubes, or stronger cables, or by develbrake shoes. The drum is bolted to the center of the wheel on the
inside. The drum rotates with the wheel, but the shoes do not. The
oping more heat-resistant pads or shoes or a cooling system;
shoes are lined with asbestos or some other material that can with(c) Effectiveness could be increased by injecting the brake
stand heat generated by friction. When the brake shoes press against
fluid into the tubes or using faster moving fluid, by using a
the drum, both the drum and the wheel stop or slow down.
more friction-sensitive shoe, or by decreasing the distance
Note. Adapted from The World Book Encyclopedia (Vol. 2, p. 463), between the shoe and the pad; (d) Brakes may malfunction
because of lack of fluid, worn or loose pads, a jammed piston
1987, Chicago: World Book, Inc. Copyright 1990 by World Book,
Inc. by permission of the publisher.
or holes in the piston, a hole in the tube or a break in the
cable line, or moisture between pads and shoes; (e) Pumping
the brakes reduces the build up of heat and allows the pad to
they would not fail? (c) What could be done to make brakes more
press on the shoe at more than one place.
effective, that is, to reduce the distance needed to stop? (d) Suppose
that you press on the brake pedal in your car but the brakes don't
In analyzing the results, Predictions 1-6 (discussed previwork. What could have gone wrong? (e) What happens when you
ously) were tested. Predictions 1 and 2 examined the types of
pump the brakes (i.e., press the pedal and release the pedal repeatedly
tests that are indicators of effective illustrations; Predictions
and rapidly)?
The verbatim retention posttest consisted of eight pairs of sentences. For each pair, one of the sentences had occurred verbatim in
the text whereas the other was a reworded version that retained the
original meaning. For example, one item on the verbatim recognition
posttest was: "Each car on a train has its own tank of compressed air.
On trains, each car has its own tank of compressed air." The instructions at the top of the sheet asked the student to "place a check mark
next to the sentence that is exactly (word-for-word) identical to a
sentence in the passage you read."
Procedure. Students were randomly assigned to treatment groups.
First, the student filled out the subject questionnaire. Second, the
student was given 8 minutes to read a booklet corresponding to his
or her treatment group. Third, the student was allowed 10 minutes
to take the recall posttest. Fourth, the student was allowed 12.5
minutes to take the transfer posttest; under the experimenter's instructions the subjects spent 2.S minutes on each item and were not
allowed to go back to previous items. Finally, the students were
allowed 3 minutes to complete the verbatim retention posttest.
Results and Discussion
Scoring. Students who rated their mechanical experience
as "very little" and who indicated having had two or fewer of
the mechanical experiences listed on the subject questionnaire
were counted as low-knowledge students. Students who rated
their mechanical experience as more than "very little" and
indicated having had more than two mechanical experiences
were counted as high-knowledge students.
For each student, four major dependent measured were
recorded: number of explanative idea units recalled (out of
35 possible); number of nonexplanative idea units recalled
(out of 60 possible); number of correct answers on the problem-solving test (out of 15 possible); and number of correct
answers on the verbatim recognition test (out of 8 possible).
Low PriDr-Knowlfitipe I earners
•
0
0
•
Control
Paris
Steps
Parts & Steps
PI
99
99
99.
99.
99.
High Prior-Knowledge Learners
• Control
E Pans
Q Steps
• Parts & Steps
vv
94
94
94
Conceptual
Recall
Nonconceplual
Recall
Problem
Solving
94
^2
Verbatim
Retention
Figure 4. Proportion correct on four posttests by treatment group
for low and high prior-knowledge learners in Experiment 1.
720
RICHARD E. MAYER AND JOAN K. GALLINI
3 and 4 assessed the types of illustrations that lead to improved
performance; and Predictions 5 and 6 focused on how the
types of learners relate to the effectiveness of illustrations.
Prediction 1: Explanative illustrations improve recall of
explanative information but not nonexplanative recall. The
first prediction is that low prior-knowledge students who read
explanative text including explanative illustrations (i.e., partsand-steps illustrations) will recall more explanative information but not more nonexplanative information as compared
to low prior-knowledge students who read text without illustrations. The top-left portion of Figure 4 shows the proportion
correct response for low prior-knowledge students in each
treatment group on explanative and nonexplanative recall.
As can be seen, the parts-and-steps group outperformed the
control group on recall of explanative information (35%
versus 15%) but not on recall of nonexplanative information
(19% versus 20%). Consistent with Prediction 1 and the
results of Mayer (1989b), separate t- tests conducted on these
data revealed that the groups differed in recall of explanative
information, t{22) « 4.53, p < .001, but not in recall of
nonexplanative information, t < 1.
Prediction 2: Explanative illustrations improve creative
problem solving but not verbatim retention. The second prediction is that low prior-knowledge students who read explanative text including explanative illustrations (i.e., both parts
and steps) will generate more creative answers to transfer
problems but will not perform better on verbatim retention
of sentence wording as compared to low prior-knowledge
students who read the text without illustrations. The top-right
portion of Figure 4 shows the proportion correct response for
low prior-knowledge students in each treatment group on
problem solving and verbatim retention. As can be seen, the
parts-and-steps group outperformed the control group on
problem solving (43% versus 24%) but not on verbatim
retention (40% versus 45%). Consistent with Prediction 2 and
the results of Mayer (1989b), separate t- tests conducted on
these data revealed the groups differed in problem solving
performance ;(22) = 4.17. p < .001, but not in verbatim
retention, t < 1.
Prediction 3: Explanative illustrations are more effective in
improving conceptual recall than nonexplanative illustrations. The third prediction is that low prior-knowledge students who read explanative text containing explanative illustrations (i.e., parts-and-steps) will show increases in explanative recall over the control group, but low prior-knowledge
students who read explanative text with static illustrations
(i.e., steps or parts) will not differ from the control group. The
top-left portion of Figure 4 shows that the parts-and-steps
group outperformed the steps, parts, and control groups on
recall of explanative information (35%, 21%, 17%, and 15%,
respectively). Consistent with Prediction 3, an analysis of
variance conducted on these data revealed the groups differed
in recall of explanative information, F(3, 44) - 9.81, MSC —
.33, p < .001, and supplemental Dunnett's Tests (at p < .05)
revealed that the parts-and-steps group outperformed the
control group, but the other illustrations groups did not
outperform the control group.
Prediction 4: Explanative illustrations are more effective in
improving problem solving performance than nonexplanative
illustrations. The fourth prediction is that low prior-knowledge students who read explanative text containing explanative illustrations (i.e., parts-and-steps) will show increases in
problem solving over the control group, but low prior-knowledge students who read explanative text with static illustrations (i.e., steps or parts) will not differ from the control
group. The top-right portion of Figure 4 shows that the partsand-steps group outperformed the steps, parts, and control
groups on problem-solving (43%, 26%, 26%, and 24%, respectively). Consistent with Prediction 4, an analysis of variance conducted on these data revealed that the groups differed
in problem-solving performance, F(3, 44) = 5.78, MSe =.31,
p < .002, and supplemental Dunnett's Tests (at p < .05)
revealed that the parts-and-steps group outperformed the
control group, but the other illustrations groups did not
outperform the control group.
Prediction 5: Explanative illustrations improve recall of
explanative information for low-knowledge students but not
for high-knowledge students. The fifth prediction is that the
strong pattern of performance on explanative recall obtained
for low-prior knowledge students (described under Prediction
3) will be attenuated or eliminated for high-knowledge students. The bottom-left panel of Figure 4 shows the proportion
correct response for high prior-knowledge students in each
treatment group on explanative recall. As can be seen, for
high prior-knowledge students, the parts-and-steps, steps,
parts, and control groups do not seem to differ in explanative
recall (28%, 34%, 30%, and 32%, respectively). Consistent
with Prediction 5, an analysis of variance conducted on these
data for high prior-knowledge students revealed no significant
differences among the groups, F(3, 44) = .61, MSC = .52, ns\
in contrast, as described under Prediction 3, the explanative
illustrations were effective in improving explanative recall for
low prior-knowledge students.
Prediction 6: Explanative illustrations improve creative
problem solving for low-knowledge students but not for highknowledge learners. The sixth prediction is that the strong
pattern of performance on problem solving obtained for low
prior-knowledge students (described under Prediction 4) will
be attenuated or eliminated for high-knowledge students. The
bottom-right panel of Figure 4 shows the proportion correct
response for high prior-knowledge students in each treatment
group on problem solving. As can be seen, for high priorknowledge students, the parts-and-steps, steps, parts, and control groups do not seem to differ in problem solving (45%,
43%, 38%, and 39%, respectively). Consistent with Prediction
6, an analysis of variance conducted on these data for high
prior-knowledge students revealed no significant differences
among the groups F(3, 44) = 1.17, MSe = .22, ns; in contrast,
as described under Prediction 4, the explanative illustrations
were effective in improving problem solving for low priorknowledge students.
Experiment 2 (Pumps)
Experiment 2 used a passage on how pumps work and was
intended to provide replicative tests of the six predictions.
WHAT MAKES A GOOD ILLUSTRATION?
Method
Subjects and design. The subjects were 96 college students recruited from the same subject pool as in Experiment 1. The design
and cell sizes were identical to Experiment 1, except that high and
low prior-knowledge status was based on students' experience with
household repair rather than car mechanics.
Materials. Similar to Experiment 1, the materials consisted of
four versions of a booklet on pumps, a subject questionnaire, and
three posttests, each typed on 8.5 x 11 inch sheets of paper, The text
was taken from the World Book Encyclopedia's (1987) entry for
"pump," containing 812 words and 128 idea units. The text included
explanations of how centrifugal, sliding vane, lift, and bicycle tire
pumps operate. No illustrations, parts illustrations, steps illustrations,
and parts-and-steps illustrations booklets were constructed as in Experiment 1. A portion of the text is given in Table 2 and examples of
each type of illustration are given in Figure 2.
The subject questionnaire was similar to that used in Experiment
1 except that it solicited information concerning the students' experience with household repair rather than automobile mechanics. For
example, students were asked to use a 5-point scale ranging from
"very little" to "very much" for the item: "Please put a check mark
indicating your knowledge of how to fix household appliances and
machines." In another item, students were asked: "Please place a
check mark next to the things you have done." The list included
owning a screw driver, owning a power saw, having replaced the
heads on a lawn sprinkler system, having replaced the washer in a
sink faucet, having replaced the flush mechanism in a toilet, and
having installed plumbing pipes or fixtures. The recall posttest was
identical to the one used in Experiment 1.
As in Experiment 1, the problem-solving posttest consisted of five
questions, each on a separate sheet. Thefivequestions were: (a) What
could be done to make a pump more reliable, that is, to make sure it
would not fail? (b) What could be done to make a pump more
effective, that is, to move more liquid or gas more rapidly? (c) Suppose
you push down and pull up the handle of a lift pump several times
but no water comes out. What could have gone wrong? (d) Why does
water enter a lift pump? Why does water exit from a lift pump? (e)
The text you read mentioned a "screw pump that consisted of a screw
rotating in a cylinder," but the text did not really explain how it
works. Based on your understanding of how pumps work, please
write your own idea of how you think a screw pump could be used
to move water.
The verbatim retention posttest corresponded to that used in
Experiment 1 except that the eight sentence pairs were based on the
pumps text. For example, one item on the verbatim recognition
721
posttest was: "Most pumps are made of steel, but some are made of
glass or plastic. Usually, pumps are made of steel, but sometimes they
are made of glass or plastic."
Procedure. The procedure was identical to that used in Experiment 1 except that students had 10 minutes to read their booklets.
Results and Discussion
Scoring. Students who rated their knowledge of household
repair as "very little" and who indicated having had two or
fewer of the experiences listed on the subject questionnaire
were counted as low prior-knowledge students; students who
rated their household-repair knowledge as more than "very
little" and indicated having had more than two householdrepair experiences were counted as high prior-knowledge students.
As in Experiment 1, four major dependent measures were
recorded for each subject: number of explanative idea units
recalled (out of 41 possible); number of nonexplanative idea
units recalled (out of 87 possible); number of correct answers
on the problem-solving test (out of 15 possible); and number
of correct answers on the verbatim recognition test (out of 8
possible). The recall, problem-solving, and verbatim retention
tests were scored as in Experiment 1. Examples of acceptable
answers for the five problem-solving questions are: (a) Reliability could be increased by using airtight seals on valves, by
using a backup system, and by using roller bearings to help
the impeller turn more easily; (b) Effectiveness could be
increased by turning the impeller faster, using larger intake
and output pipes, and increasing the diameter of the cylinder;
(c) A lift pump might malfunction because there is no water
in the pump, a valve is stuck, a seal is broken, a blockage has
formed in the line, and the piston has become unattached
from the handle; (d) Water enters a lift pump because of
suction or a vacuum, whereas water exits because of compression or pressure, and (3) A screw pump has a rotating screw
with threads that seals tightly against the inside of the cylinder.
In analyzing the results, we tested the same six predictions
as in Experiment 1.
Prediction I: Explanative illustrations improve recall of
explanative information but not nonexplanative recall. The
top-left portion of Figure 5 shows the proportion correct
response for low prior-knowledge students in each treatment
group on explanative and nonexplanative recall. As can be
Table 2
seen, the parts-and-steps group outperformed the control
Portion of Text on Pumps (With Explanative
group
on recall of explanative information (23% versus 3%)
Information in Italics)
but not on recall of nonexplanative information (9% versus
16%). Consistent with Prediction 1, the results of Experiment
Bicycle tire pumps vary in the number and location of the valves
I, and the results of Mayer (1989b), separate t- tests conducted
they have and in the way air enters the cylinder. Some simple bicycle
tire pumps have the inlet valve on the piston and the outlet valve at
on these data revealed that the parts-and-steps group outperthe closed end of the cylinder. A bicycle tire pump has a piston that
formed the control group in recall of explanative information,
moves up and down. Air enters the pump near the point where the
t(28) = 4.76, p < .001. In addition, the parts-and-steps group
connecting rod passes through the cylinder. As the rod is pulled out,
not only failed to outperform the control group on recall of
air passes through the piston and fills the areas between the piston
and the outlet valve. As the rod is pushed in, the inlet valve closes and nonexplanative information as in Experiment 1; in Experithe piston forces air through the outlet valve.
ment 2, the control group actually recalled significantly more
nonexplanative information than the control group, Z(28) =
Note. Adapted from The World Book Encyclopedia (Vol. 15, p.
4.19, p < .001, as would be predicted by a strong version of
794), 1987, Chicago: World Book, Inc. Copyright 1990 by World
schema theory.
Book, Inc. by permission of the publisher.
722
RICHARD E. MAYER AND JOAN K. GALLINI
Low Prior-Knowledge Learners
u ,5-
•
E2
Q
H
Control
Paris
Steps
Parts & Steps
99
99
9%
9v
9
9V
0
9
9y
v
99
99
99
9i
o .3-
I
.2-
Concapiual
Recall
9V
9
v
9
y
9
9y
Nonconceptual
Recall
Problem
Solving
J
High Prior-Knowledge Learners
.6.5u
Q
0
0
•
Verbatim
Releniion
Control
Parts
Steps
Pans & Steps
%
94
90
9
v
9
v
9%
90
9
y
9
v
9
v
9
0
9
9
>
9%
9v
99
.4-
o .3.2-
.1 •
Conceptual
Recall
Nonconceptual
Recall
Problem
Solving
Verbalim
Retention
Figure 5. Proportion correct on four posttests by treatment group
for low and high prior-knowledge learners in Experiment 2.
Prediction 2: Explanative illustrations improve creative
problem solving but not verbatim retention. The top-right
portion of Figure 5 shows the proportion correct response for
low prior-knowledge students in each treatment group on
problem solving and verbatim retention. As can be seen, the
parts-and-steps group outperformed the control group on
problem solving (47% versus 28%) but not on verbatim
retention (66% versus 68%). Consistent with Prediction 2, the
results of Experiment 1, and the results of Mayer (1989b),
separate t- tests conducted on these data revealed that the
groups differed in problem solving performance, /(28) = 3.51,
p < .01, but not in verbatim retention, /(28) < 1.
Prediction 3: Explanative illustrations are more effective in
improving conceptual recall than nonexplanative illustrations. The top-left portion of Figure 5 shows that the partsand-steps group outperformed the steps, parts, and control
groups on recall of explanative information (23%, 13%, 13%,
and 3%, respectively). Consistent with Prediction 3 and the
results of Experiment 1, an analysis of variance conducted on
these data revealed the groups differed in recall of explanative
information, F(3, 56) = 7.32, MSe = .53, p < .001, and
supplemental Dunnetf s Tests (at p < .05) revealed that the
parts-and-steps group outperformed the control group, but
the other illustrations groups did not outperform the control
group.
Prediction 4: Explanative illustrations are more effective in
improving problem solving performance than nonexplanative
illustrations. The top-right portion of Figure 5 shows that
the parts-and-steps group outperformed the steps, parts, and
control groups on problem solving (47%, 20%, 28%, and
28%, respectively). Consistent with Prediction 4 and the results of Experiment 1, an analysis of variance conducted on
these data revealed the groups differed in problem-solving
performance, F(3, 56) = 9.20, MSe = .31, p < .001, and
supplemental Dunnett's Tests (at p < .05) revealed that the
parts-and-steps group outperformed the control group, but
the other illustrations groups did not outperform the control
group.
Prediction 5: Explanative illustrations improve recall of
explanative information for low-knowledge students but not
for h igh-knowledge students. The bottom-left panel of Figure
5 shows the proportion correct response for high prior-knowledge students in each treatment group on explanative recall.
As can be seen, for high prior-knowledge students, the differences among the parts-and-steps, steps, parts, and control
groups in explanative recall (24%, 23%, 16%, and 13%,
respectively) were relatively small. However, in contrast to
Prediction 5 and the results of Experiment 1, an analysis of
variance conducted on these data for high prior-knowledge
students revealed mildly significant differences among the
groups, F(3, 56) = 2.98, MSC = .53, p < .05. Apparently, the
emphasis on explanative information in the illustrations influenced the recall performance of high prior-knowledge in
this experiment,
Prediction 6: Explanative illustrations improve creative
problem solving for low-knowledge illustrations but not for
high know/edge learners. The bottom-right panel of Figure
5 shows the proportion correct response for high prior-knowledge students in each treatment group on problem solving.
As can be seen, for high prior-knowledge students, the partsand-steps, steps, parts, and control groups do not seem to
differ in problem solving (55%, 45%, 49%, and 51%, respectively). Consistent with Prediction 6, an analysis of variance
conducted on these data for high prior-knowledge students
revealed no significant differences among the groups, F(3, 56)
- 5.66, MSC = .72, ns\ in contrast, as described under Prediction 4, the explanative illustrations were effective in improving problem solving for low prior-knowledge students.
Experiment 3 (Generators)
Experiment 3 used a passage on electric generators and was
intended to provide replicative tests of the six predictions.
This experiment differed from Experiments 1 and 2 in several
important ways: the passage was longer and more complex,
there were only three versions of the booklet rather than four,
the parts-and-steps illustrations contained labels for parts but
not for steps, the parts-and-steps illustrations booklet included
WHAT MAKES A GOOD ILLUSTRATION?
several illustrations that were not in the same format as used
in previous experiments, words were added to the parts and
parts-and-steps booklets to clarify the illustrations, and prior
knowledge was measured on a different kind of survey instrument.
Method
Subjects and design. The subjects were 108 college students recruited from the same subject pool as Experiments 1 and 2. The
design was identical to Experiment 1 except that there was no steps
group; the no-illustrations group consisted of 17 low prior-knowledge
and 18 high prior-knowledge students; the parts group consisted of
15 low prior-knowledge and 21 high-prior knowledge students; and
the parts-and-steps group consisted of 12 low-prior knowledge and
25 high prior-knowledge students.
Materials. The materials consisted of three versions of a passage
on electric generators, a subject questionnaire, and three posttests,2
each typed on 8.5 x 11 in sheets of paper.
The text for the no-illustrations booklet was taken from the World
Book Encyclopedia's (1987) entry on "electric generator," containing
2066 words and 161 idea units. The parts-illustrations booklet was
created by inserting 4 figures and approximately 225 words that
clarified the figures. The figures showed the parts in a simplified
generator, a real-life generator, a simplified AC generator, and a reallife AC generator. The parts-and-steps illustrations booklet was created by adding 13figuresand 372 words that clarified the figures. It
included the same figures as in the parts booklet along with the
following additions: (a) three additional illustrations of the simple
generator showing the armature in each state of rotation; (b) three
additional illustrations of the simplified AC generator showing the
armature in each state of rotation; (c) two additional illustrations
showing the relation between the simplified and real-life generator
and between the simplified AC generator and real-life AC generator;
and (d) a summary illustration showing the flow of current among
the parts in a simplified AC generator.
The subject questionnaire consisted of 11 statements concerning
the subject's experience and interest in working with mechanical and
electrical devices. Each statement was to be rated on a 5-point scale
consisting of never (1), seldom (2), sometimes OX frequently (4), and
always (5). Typical questionnaire items included: "I like to repair
electrical gadgets." "I enjoy experimenting with mechanical things,"
and "I like to disassemble mechanical things merely to perceive what
they look like inside."
The recall posttest was identical to that used in Experiments 1
and 2.
The problem-solving posttest contained eight questions: (a) Why
is no energy generated when the armature of the generator is in a
parallel position? (b) Provide reasons as to why a generator might not
produce enough electrical energy, (c) What can be done to increase
the energy output from an electrical generator? (d) How would you
know if an AC electric generator is not functioning properly? (e)
Explain why one generator might produce more cycles of current
than another? (0 What happens to the electric current if no device is
hooked up to the electric generator? (g) Can an AC generator ever
conduct electric current in the same direction throughout one complete cycle of the rotation process? Explain your answer, (h) Suppose
a radio is hooked up to an AC generator. The radio suddenly goes
out. The problem seems to be related to the generator. Provide an
explanation for the power failure.
The verbatim retention posttest consisted of 15 pairs of sentences
such as, "An electric generator is a machine that produces electricity.
An electric generator is a machine that generates electricity.
723
Procedure. The procedure was the same as in Experiments 1 and
2 except that subjects were given 25 minutes to read their booklets.
Results and Discussion
Scoring. Students' scores on the subject questionnaire
were computed by tallying the numbers (i.e., 1 through 5)
circled for each of the eleven items. Students who obtained
scores below 29 were counted as low prior-knowledge students
and those who obtained scores of 29 or above were counted
as high prior-knowledge students.
As in Experiments 1 and 2, four major dependent measures
were recorded for each subject: number of explanative idea
units recalled (out of 32 possible); number of nonexplanative
idea units recalled (out of 129 possible); number of correct
answers on the problem-solving test (out of 15 possible); and
number of correct answers on the verbatim recognition test
(out of 15 possible). The recall, problem-solving, and verbatim
retention tests were scored as in Experiment 1. Examples of
acceptable answers for the problem-solving questions are: (a)
No energy is generated when the armature is in a parallel
position because it is not cutting through the magnetic lines
of force; (b) The generator might not produce enough electrical energy because the carbon brushes are not properly aligned
with the slip ring, or the initial current from the prime mover
is insufficient; (c) One sign of an AC generator malfunctioning
is that the current does not reverse direction; (d) A generator
might produce more cycles of current than another because
it is larger or the armature is rotating at a faster rate; (e) If no
device is hooked up to an electric generator, no electrical
current is produced; (f) A generator that conducts electrical
current in the same direction is not an AC generator but a
DC generator; and (g) If a radio hooked up to a generator
suddenly fails, the problem could be that the armature stopped
rotating or the slip rings are not rotating properly.
As in Experiments 1 and 2, six predictions were tested.
Prediction 1: Explanative illustrations improve recall of
explanative information but not nonexplanative recall. The
top-left portion of Figure 6 shows the proportion correct
response for low prior-knowledge students in each treatment
group on explanative and nonexplanative recall. As can be
seen, the parts-and-steps group outperformed the control
group on recall of explanative information (21% versus 9%)
but not on recall of nonexplanative information (7% versus
8%). Consistent with Prediction 1, the results of Experiments
1 and 2, and the results of Mayer (1989b), separate t-tests
conducted on these data revealed that the groups differed in
recall of explanative information, t(27) = 3.37, p < .01, but
not in recall of nonexplanative information, /(27) < 1, ns.
Prediction 2: Explanative illustrations improve creative
problem solving but not verbatim retention. The top-right
portion of Figure 6 shows the proportion correct response for
low prior-knowledge students in each treatment group on
problem solving and verbatim retention. As can be seen, the
parts-and-steps group outperformed the control group on
2
Several additional posttests were given but were not analyzed
because they were not relevant to our six predictions.
724
RICHARD E. MAYER AND JOAN K. GALLINI
UQW Pripr-Knowledge Learners
D
0
|
Control
Paris
Parts & Steps
Conceptual
Recall
Nonconceptjal
Recall
Problem
Solving
Verbatim
Retention
Problem
Solving
Verbatim
Retention
High Prior-Knowledge Learners
Q
Control
E3 Parts
• Paris & Sieps
S -3.2-
Conceptual
Recall
Nonconceplual
Recall
Figure 6. Proportion correct on four posttests by treatment group
for low and high prior-knowledge learners in Experiment 3.
problem solving (56% versus 29%) but not on verbatim
retention (67% versus 75%). Consistent with Prediction 2, the
results of Experiments 1 and 2, and the results of Mayer
(1989b), separate t- tests conducted on these data revealed
that the groups differed in problem-solving performance t(21)
= 4.59, p < .001, but not in verbatim retention, r(27) = 1.32,
ns.
Prediction 3: Exploitative illustrations are more effective in
improving conceptual recall than nonexplanative illustrations. The top-left portion of Figure 6 shows that the partsand-steps group outperformed the parts and control groups
on recall of explanative information (21%, 12%, and 9%,
respectively). Consistent with Prediction 3, an analysis of
variance conducted on these data revealed that the groups
differed in recall of explanative information, F(2, 40) =
11.348, p < .001, and supplemental Dunnett's Tests (at p <
.05) revealed that the parts-and-steps group outperformed the
control group, but the parts groups did not outperform the
control group.
Prediction 4: Explanative illustrations are more effective in
improving problem solving performance than nonexplanative
illustrations. The top-right portion of Figure 6 shows that
the parts-and-steps group outperformed the parts and control
groups on problem solving (56%, 39%, and 29%, respectively). Consistent with Prediction 4, an analysis of variance
conducted on these data revealed that the groups differed in
problem-solving performance F(2, 40) = 21.1 \,p< .001,and
supplemental Dunnett's Tests (at p < .05) revealed that the
parts-and-steps group outperformed the control group, but
the parts group did not outperform the control group.
Prediction 5: Explanative illustrations improve recall of
explanative information for low-knowledge students but not
for high-knowledge students. The bottom-left panel of Figure
6 shows the proportion correct response for high prior-knowledge students in each treatment group on explanative recall.
As can be seen, for high prior-knowledge students, the differences among the parts-and-steps, parts, and control groups in
explanative recall (26%, 16%, and 11%, respectively) were
relatively large. Inconsistent with Prediction 5 and the results
of Experiments 1 and 2, an analysis of variance conducted on
these data for high prior-knowledge students showed significant differences among the groups, F(2, 60) = 19.96, p <
.001, and is similar to the pattern for low prior-knowledge
students described under Prediction 3 in which explanative
illustrations were effective in improving explanative recall.
Thus, both in Experiments 2 and 3, this is the only inconsistency in our predictions.
Prediction 6: Explanative illustrations improve creative
problem solving for low-knowledge students but not for highknowledge learners. The bottom-right panel of Figure 6
shows the proportion correct response for high prior-knowledge students in each treatment group on problem solving.
As can be seen, for high prior-knowledge students, the partsand-steps, parts, and control groups do not seem to differ in
problem solving (61%, 55%, and 55%, respectively). Consistent with Prediction 6 and the results of Experiments 1 and 2,
an analysis of variance conducted on these data for high priorknowledge students revealed no significant differences among
the groups, F(2,60) = 1.20, ns; in contrast, as described under
Prediction 4, the explanative illustrations were effective in
improving problem solving for low prior-knowledge students.
Conclusion
Conditions for Effective Illustrations
A general goal of this study was to test the model of
conditions for effective illustrations summarized in Figure 3.
We examined the specific predictions that illustrations would
be effective in our experiments only with explanative text
(i.e., appropriate text), tests of understanding and reasoning
(i.e., appropriate tests), steps-and-parts illustrations (i.e., appropriate illustrations), and low prior-knowledge learners (i.e.,
appropriate learners). Table 3 summarizes whether or not
each of the six predictions, based on the model of conditions
for effective illustrations, was supported in each of the three
experiments.
725
WHAT MAKES A GOOD ILLUSTRATION?
Table 3
Summary of Results: Tests of18 Predictions
Results of Experiment
Predictions based on four conditions for effective illustrations (in boldface)
Appropriate text: Use exploitative text
(No predictions were tested.)
Appropriate test: Use tests that measure conceptual understanding
Explanative illustrations improve explanative recall but not nonexplanative
recall.
Explanative illustrations improve creative problem solving but not verbatim
retention.
Appropriate illustrations: Use illustrations that explain
Explanative illustrations improve explanative recall better than nonexplanative
illustrations.
Explanative illustrations improve problem solving better than nonexplanative
illustrations.
Appropriate learners: Use illustrations for learners who lack prior knowledge
Explanative illustrations improve explanative recall for low rather than high
knowledge learners.
Explanative illustrations improve problem-solving for low rather than high
knowledge learners.
Note. Y = prediction was confirmed', N = prediction was not confirmed.
Appropriate text. The first condition in the model is that
the text be appropriate for the instructional goal. Since our
instructional goal was to enhance learner understanding, we
used explanative text rather than descriptive or narrative text.
Although we did not directly manipulate the type of text in
this study, we exclusively used text in each experiment that
met the condition of appropriateness.
Appropriate tests. The second condition is that the test be
appropriate to the instructional goal. In our experiments we
evaluated student performance on appropriate tests, such as
explanative recall and problem solving, and on inappropriate
tests, such as nonexplanative recall and verbatim retention.
The first two predictions in Table 3, relating to test appropriateness, were supported in each of the three experiments. The
consistency of these results, in conjunction with similar results
in two previous experiments (Mayer, 1989b), provide strong
support for the idea that tests should be sensitive to the specific
goals of instruction.
Appropriate illustrations. The third condition is that the
illustrations should be appropriate for the instructional goal.
In our experiments, we compared illustrations that concretized the changes in status of parts within the system (explanative illustrations) to illustrations that did not (nonexplanative
illustrations). The second two predictions in Table 3, concerning illustration appropriateness, were supported in each of the
three experiments. These results provide new evidence concerning the characteristics of the models portrayed in effective
explanative illustrations: The ineffective illustrations failed to
visually portray either system topology (i.e., steps illustrations)
or component behavior (i.e., parts illustrations) whereas the
effective illustrations (i.e., parts-and-steps illustrations) portrayed both. In particular, the parts-and-steps illustrations
used a series of two (or more) frames to show the state of the
components within the system at various points in the operation of the system.
Appropriate learners. The fourth condition is that the
learners would not achieve the desired learning outcome
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
N
N
Y
Y
Y
without special instruction. In our experiments, we compared
the learning outcomes of students who lacked domain-specific
knowledge and those who possessed it. The final two predictions in Table 3, testing learner appropriateness, were generally supported. The major exception, that explanative illiistrations improved explanative recall for high-knowledge learners
in Experiment 3, may be accounted for by the fact that the
explanative text was expanded upon in Experiment 3 but not
in the other experiments.
In summary, in the spirit of complementing Larkin &
Simon's (1987) work, our goal was to answer the question,
"When is a diagram worth ten thousand words?" Our results
provide a four-part answer: when the text is potentially understandable, when the value of illustrations is measured in
terms of learner understanding, when the illustrations explain,
and when the student lacks previous experience. Finally, our
report fits into a growing literature on the cognitive psychology of visual instructional methods (Holley & Dansereau,
1984; Waddill, McDaniel & Einstein, 1988; Weidenmann,
1989; Winn, 1987) and points to the potential of visually
based instruction as a medium for promoting students' understanding of scientific material.
References
Alexander, P. A., & Judy, J. E. (1988). The interaction of domainspecific and strategic knowledge in academic performance. Review
of Educational Research, 58, 375-404.
Bobrow, D. G. (Ed.). (1985). Qualitative reasoning about physical
systems. Cambridge, MA: MIT Press.
de Kleer, J., & Brown, J. S. (1983). Assumptions and ambiguities in
mechanistic mental models. In D. Gentner & A. L. Stevens (Eds.),
Mental models (pp. 172-190). Hillsdale, NJ: Erlbaum.
de Kleer, J., &. Brown, J. S. (1985). A qualitative physics based on
confluences. In D. G. Bobrow (Ed.), Qualitative reasoning about
physical systems (pp. 7-83). Cambridge, MA: MIT Press.
Gentner, D., & Gentner, D. R. (1983). Flowing waters or teeming
crowds: Mental models of electricity. In D. Gentner & A. L. Stevens
726
RICHARD E. MAYER AND JOAN K. GALLINI
(Eds.), Mental models (pp. 101-130). Hillsdale, NJ: Erlbaum.
Gentner, D., & Stevens, A. L. (Eds.). (1983). Mental models. Hillsdale,
NJ: Erlbaum.
Holley, C. D., & Dansereau? D. F. (Eds.). (1984). Spatial learning
strategies. San Diego, CA: Academic Press.
Kieras, D. E., & Bovair, S. (1984). The role of a mental model in
learning to operate a device. Cognitive Science, 8, 255-274.
Larkin, J. H., & Simon, H. A. (1987). Why a diagram is (sometimes)
worth ten thousand words. Cognitive Science, 11, 65-99.
Levie, W. H., Sc Lentz, R. (1982). Effects of text illustrations: A
review of research. Educational Communication and Technology
Journal, SO, 195-232.
Levin, J. R. (1981). On the functions of pictures in prose. In F. J.
Pirozzolo & M. C. Wittrock (Eds.), Neuropsychological and cognitive processes in reading {pp. 203-228). San Diego, CA: Academic
Press.
Levin, J. R., Anglin, G. J., & Carney, R. N. (1987). On empirically
validating functions of pictures in prose. In D. M. Willows Sc H.
A. Houghton (Eds.), The psychology of illustration: Vol. 1. Basic
research (pp. 51-85). New York: Springer-Verlag.
Mandl, H., & Levin, J. R. (Eds.). (1989). Knowledge acquisition from
text and pictures. Amsterdam: Elsevier.
Mayer, R. E. (1984). Aids to text comprehension. Educational Psychologist, 19, 30-42.
Mayer, R. E. (1985). How to analyze science prose. In B. K. Britton
& J. B. Black (Eds.), Understanding expository text (pp. 305-313).
Hillsdale, NJ: Erlbaum.
Mayer, R. E. (1989a). Models for understanding. Review of Educational Research, 59, 43-64.
Mayer, R. E. (1989b). Systematic thinking fostered by illustrations in
scientific text. Journal of Educational Psychology, 81, 240-246.
McDaniel, M. A., & Pressley, M. (1987). Imagery and related mnemonic processes. New York: Springer-Verlag.
Rumelhart, D. E., & Ortony, A. (1977). The representation of knowledge in memory. In R., C. Anderson, R. J. Spiro, & W. E. Montague
(Eds.), Schooling and the acquisition of knowledge (pp. 99-135).
Hillsdale, NJ: Erlbaum.
Waddill, P. J., McDaniel, M. A., & Einstein, G. O. (1988). Illustrations as adjuncts to prose: A text-appropriate processing approach.
Journal of Educational Psychology, 80, 457-464.
Weidenmann, B. (1989). When good pictures fail: An informationprocessing approach to the effects of illustrations. In H. Mandl &
J. R. Levin (Eds.), Knowledge acquisition from text and pictures
(pp. 157-171). Amsterdam: Elsevier.
White, B. Y., & Frederiksen, J. R. (1987). Qualitative models and
intelligent learning environments. In R. W. Lawler & M. Yazdani
(Eds.), Artificial intelligence and education: Vol. 1. Learning environments and tutoring systems (pp. 281-305). Norwood, NJ: Ablex.
Willows, D. M., & Houghton, H. A. (Eds.). (1987). The psychology
of illustration: Vol. 1. Basic research. New York: Springer-Verlag.
Winn, B. (1987). Charts, graphs, and diagrams in educational materials. In D. M. Willows & H. A. Houghton (Eds.), The psychology
of illustration: Vol. 1. Basic research (pp. 152-198). New York:
Springer-Verlag.
The world book encyclopedia. (1987). Chicago: World Book.
Received September 11, 1989
Revision received May 11, 1990
Accepted May 29, 1990