In the current study the level of realism in visualizations and the role of diverse presentation formats of dynamic and different static visualizations in a complex, dynamic domain (locomotion pattern classification) were investigated. In a two-by-three design with the two independent factors realism (realistic, schematic) and presentation format (dynamic, static-sequential, static-simultaneous) one hundred-and-twenty university students were randomly assigned to six conditions. Learners had to learn how to classify fish according to their locomotion pattern. Learning outcomes were measured by two pictorial tests, assessing recognition and transfer performance. Data analyses showed an advantage of the dynamic conditions over the sequential conditions in both recognition and transfer performance. Simultaneous visualizations did not lead to different learning outcomes than either dynamic or sequential visualizations. Moreover, there was no main effect for realism or an interaction of realism with presentation format. Implications for the design of instructional materials are discussed.