This paper provides experimental and computational support for the hypothesis that categorization is accomplished by modifying the similarity space underlying perception by investigating the change in the initial structure of perceived similarity of objects after perceptual learning and categorization. Data from two human experiments on discrimination and categorization of multi-dimensional stimuli were modeled with a connectionist model, in order to determine how the initial structure of perceived similarity between stimuli influences the final structure of the similarity space after categorization. An interesting finding was that, on the one hand, categorization based on a salient dimension may improve discrimination along dimensions that are not salient for the task, and, on the other hand, categorization based on a non-salient dimension may still reveal separation along salient dimensions. The results of the reported study may provide additional insights into a possible mechanism of categorization. 1 Theoretical overview 1.1