The authors propose and test an exemplar-based random-walk model for predicting response times in tasks of speeded, multidimensional perceptual classification. The model combines elements of Nosofsky's (1986) generalized context model of categorization and Logan's (1988) instance-based model of automaticity. In the model, exemplars race among one another to be retrieved from memory with rates determined by their similarity to test items; the retrieved exemplars provide incremental information that enters into a random walk process. The model predicts correctly effects of within- and between- category similarity; effects of individual-object familiarity; and effects of extended practice on classification response times. It also builds bridges between the domains of categorization and automaticity.