Current initiatives to transform undergraduate STEM education in the United States advocate for the use of multidimensional learning, wherein instruction provides opportunities for students to gain competency with fundamental or cross-disciplinary concepts and practices. To achieve this goal, the discipline-based education research (DBER) community is drawing on theories and approaches from the cognitive sciences to better understand how instructional choices may influence student cognition and performance. In this study, we investigate the extent to which adding visual representation to exam questions may alter the cognitive demand placed on the learner, thereby affecting exam performance. Exam questions were crafted in pairs -- one form embedded some necessary information within a visual representation, while the other comprised only text -- and distributed on exams across an undergraduate molecular life sciences curriculum. Comparison of analogous questions indicates that visual representation does affect performance on most questions; however, the nature of that effect depends on other features (e.g., format, cognitive level) of the task. Adding a visual representation to difficult open-response questions often decreased performance further, while adding a representation to similarly difficult forced-response items was more likely to increase performance. Drawing on cognitive theories of learning, we rationalize how the presence and interactivity of these elements may affect the cognitive load of the task and working memory efficiency. The findings of this study have implications for instructors regarding the design and interpretation of student assessments and call on researchers for deeper investigation of the relationship between student cognition and multidimensional assessments.