The theme of optimal performance has always played a central role in theorizing involving Nosofsky's (1986) generalized context model, a well known exemplar model of category learning and representation. In this chapter I review and elucidate this theme. First, there is support for the hypothesis that observers may learn to distribute their attention over psychological dimensions so as to nearly optimize performance in any given task. This attention-optimization hypothesis has allowed the exemplar model to accurately describe relations among categorization and other fundamental cognitive processes in a wide variety of experimental paradigms. Second, the exemplar model can be given an interpretation as a likelihood-based model of categorization decision making. The model therefore predicts that, regardless of the complexity of the category structure, observers will often learn category decision boundaries with a nearly optimal form. Experiments are reviewed that test this prediction of the exemplar model against alternative models that place constraints on the functional forms of category decision boundaries. Finally, I review recent developments that have led to the exemplar model being endowed with a more optimal response-selection mechanism.