Examined the hypothesis that information enhancing category differences receives greater weight in estimates of category means than information that reduces such differences. In the first experiment, subjects estimated the cumulative means of test scores of two groups of students. The experimental manipulation involved a gradual shift of the true mean of one group either towards or away from the true mean of the other group. As predicted, changes of estimates were larger when the two means became more dissimilar than when they became more similar. The second experiment involved otherwise identical procedures, but the variance in one category was increased while the mean remained stable. Subjects perceived an illusory change of the mean away from the comparison category. It is suggested that accentuation effects of this kind may limit the reduction of social stereotypes.