This experiment (N = 239) investigated the effects of group power and legitimacy of power differentials on intergroup discrimination, measured through negative outcome allocations and linguistic abstraction. Furthermore, it examined the mechanisms through which group power affects discrimination by testing the mediating role of perceived interdependence and social identification. Three power conditions were created by modifying the standard minimal group paradigm: equal, high and low power conditions. Power was directly proportional (legitimate conditions) or inversely proportional (illegitimate conditions) to group members' performance in a problem-solving task. Results showed that intergroup discrimination in the high and low power conditions was higher than in the equal power condition on both the allocation and the linguistic measures. Legitimacy moderated the effect of group power on negative outcome allocations, while it had a main effect on the linguistic abstraction. In addition, perceived interdependence turned out to be the main mediator of the effect of power on negative outcome allocations. Copyright © 2006 John Wiley & Sons, Ltd.