Many theories of concepts link categorizing to similarity. If a new instance is sufficiently similar to category members, then the instance is likely to be a member itself. However, judged similarity and judged category likelihood sometimes diverge. In these studies, we describe frequency distributions for categories that vary along a single dimension, and ask Ss to rate the similarity, typicality, or category likelihood of instances along this continuum. The average ratings exhibit distinct patterns, with category likelihood depending on the instance's frequency and with similarity depending on distance from the instance to the center of the distribution. Typicality ratings show effects of both frequency and distance. These differences occur for bimodal distributions (Experiments 1 and 2) and for unimodal ones (Experiment 3). They appear both when we present the distributions as histograms and when we imply them in descriptions. We argue that similarity-based models of categorizing are incomplete and may apply mainly to situations in which more definitive information is unavailable.