We argue that representations of infrequently encountered groups are more likely to be congruent with stereotype-based expectations than perceptions of more frequently encountered groups. This is because it should be easier to maintain stereotype-based expectations in view of a limited number of expectancy-confirming and disconfirming instances than in view of a larger body of information. Results of 2 experiments in a modified illusory correlation paradigm supported this prediction. In both experiments, participants read a set of statements about the behavior of the members of 2 groups, the 1 group occurring more frequently. Prior to this, either positive or negative expectations about the groups were induced. Experiment 2 also included a manipulation of processing load. Representations of the groups (as measured by behavioral assignments, frequency estimates, and free recall) were more expectancy-congruent for the infrequently encountered group and expectancy-incongruent for the frequently encountered group. This effect was not moderated by processing load.