A common judgmental task involves predicting the category membership of an individual on the basis of information specific to that individual and background information regarding the base rate of different categories. According to statistical theory, predictions may deviate from base rates only to the extent that the individuating information is diagnostic. Previous research has demonstrated that diagnosticity is often judged by “representativeness,” the degree to which the individuating information is differentially suggestive of the different possible categories. Thus, information with high differential representativeness will swamp base-rate information even if it is almost worthless (e.g., because its source is unreliable). The present studies varied differential representativeness by manipulating the prediction categories’ similarity to one another vis-a-vis the individuating information. It was found that the effect of the base rate increased systematically as differential representativeness decreased. Representativeness was measured independently by several converging techniques. These measures predicted the magnitude of the base-rate effect, supporting the hypothesis that neutral stimuli are assigned to categories in proportion to the base rates.