In self-directed learning tasks, participants can control the sequencing and timing of information presentation. In contrast, the existing literature on category learning has focused on passive learning paradigms, wherein information is presented to the learner randomly or in a pattern determined in advance by the experimenter. To explore the impact of self-directed learning on categorization performance, we compared passive and self-directed learning in the seminal Shepard, Hovland, and Jenkins (1961) category learning tasks. Participants learned by actively querying the category membership of individual exemplars in an array or by passively viewing labeled examples. We found that active learners exhibited significantly faster learning than passive learners. In addition, the benefits of self-directed learning were not uniform, but varied as a function of the category structure. Our results suggest that differences in interaction with learning materials can alter the difficulty of learning problems, independent of the abstract structure of the underlying rule.