Classification performance in the dot-pattern, prototype-distortion paradigm (e.g., Posner & Keele, 1968, 1970) was modeled within a multidimensional scaling (MDS) framework. Similarity-judgment data were used to derive MDS solutions for sets of dot patterns that were generated from prototypes. These MDS solutions were then used in conjunction with exemplar, prototype, and mixture models to predict classification and recognition performance. Across three experiments, an MDS-based exemplar model accounted for the main effects of a variety of fundamental learning variables, including effects of level of distortion of the patterns, category size, delay of transfer phase, and individual item frequency. Most important, the model was capable of quantitatively predicting classification and recognition probabilities for individual dot patterns in the sets, not simply general trends of performance. An MDS-based prototype model fared poorly relative to the exemplar model, and the mixed model provided little evidence for the existence of a prototype-abstraction process that operated above and beyond pure exemplar-based generalization. 2