A model of graph comprehension is proposed including perceptual and memory processes. Multidimensional scaling (MDS), cluster analysis, and analysis of variance (ANOVA) were used to determine how college students with different mathematical experience read different types of bar graphs. Data were collected at the University of Oklahoma (Norman) during the summer 1991 semester from the following subjects: 28 students (mostly freshmen) in an intermediate algebra course; 39 students (mainly sophomores) in a Calculus II course; and 28 seniors and graduate students in an upper division calculus-based applied statistics course. The MDS and cluster analysis showed that there were differences in how the different groups of subjects clustered the different bar graph types. Of four bar graph types and two orientations, the graph types clustered together were simple and grouped (side by side) bar graphs. Using ANOVA, it was found that these graph types were read equally well and were read more accurately than were the other graph types. This finding indicates that graphs of different types and levels of perceptual complexity may share a single mental model or schema for graph reading. Subjects with more mathematical experience used graph orientation to a greater degree in their typicality ratings than did subjects with less mathematical experience, and it appeared that they may have had better strategies for recognizing the trends displayed in graphs. Seven figures and 1 table illustrate the study, and there is a 27-item list of references. (Author/SLD)