Two studies examined whether infants' category discrimination in an object-examination task was based solely on an ad hoc analysis of perceptual similarities among the experimental stimuli. In Experiment 1A, 11-month-olds examined four different exemplars of one superordinate category (animals or furniture) twice, followed by a new exemplar of the familiar category and an exemplar of the contrasting category. Group A (N = 39) explored natural-looking toy replicas with low between-category similarity, whereas group B (N = 40) explored artificial-looking toy models with high between-category similarity. Experiment 1B (N = 40) tested a group of 10-month-olds with the same design. Experiment 1C (N = 20) reversed the order of test trials. For Experiment 2 (N = 20), the same artificial-looking toy animals as in Experiment 1 (group B) were used for familiarization), but no category change was introduced at the end of the session. Infants' responses varied systematically only with the presence of a category change, and not with the degree of between-category similarity. This supports the hypothesis that performance was knowledge based.