Adaptive network and exemplar-similarity models were compared on their ability to predict category learning and transfer data. An exemplar-based network that combines key aspects of both modeling approaches was also tested. The exemplar-based network incorporates an exemplar-based category representation in which exemplars become associated to categories via the same error-driven, interactive learning rules that are assumed in standard adaptive networks. Experiment 1, which involved a partial replication and extension of a probabilistic classification learning paradigm conducted previously by Gluck and Bower (1988a), demonstrated the importance of an error-driven learning rule. Experiment 2, which involved an extension of a classification learning paradigm used by Medin and Schaffer (1978) for discriminating between exemplar and prototype models, demonstrated the importance of an exemplar-based category representation. Among the models tested here, only the exemplar- based network accounted for all the major qualitative phenomena of interest across both experiments. The model also achieved good quantitative predictions of the learning and transfer data in both experiments.