Two studies investigated the pervasiveness of race as a social categorization and whether the organization of information around racial categories is sensitive to contextual factors. Both studies measured accentuation effects (more intra- than inter-race errors) and own-group bias (fewer confusions between own- than other-group members) in person memory, using the paradigm developed by Taylor, Fiske, Etcoff and Ruderman (1978). Experiment 1 studies the generalization of these effects across ethnic group membership (black/white) and topic (categorization-relevant/irrelevant) in a 2 × 2 [× 4] between-subject design, with type of error as a repeated measure. There was a highly significant accentuation effect, which was not affected by either topic or group membership. Experiment 2, using white subjects only, manipulated anticipated future interaction/no interaction, which affected overall accuracy/error rate, but not the strong accentuation effect. Neither study found any support for an own-group bias. Results are discussed in terms of the automaticity of race as a basis for social categorization.