The present research had two purposes. First, to investigate the effects of observational demands in relation to the impact that prior impressions have on the encoding and storage of subsequent performance. Second, the extent to which these encoded and stored behaviors are used to modify existing impressions. Specifically, level of observational processing demands and ratee prior performance information were manipulated in a 3 × 3 factorial design. Results indicated that prior impressions either had no effect or biased processing toward impression-inconsistent behaviors and that as observational demands increased, judgments were biased toward prior impressions. These results were discussed both in relation to the debate over assimilation and contrast effects and to the more basic issue of modifying existing impressions. © 1993 Academic Press, Inc.