Two experiments evaluated the effect of role assignment on intergroup bias. A social categorization model (Brewer & Miller, 1984) predicts a reduction in bias when the basis for assignment to task roles or subgroup composition in a contact situation is category independent (cross-cut) rather than category related (convergent). Conversely, a model based on social identity theory (Brown & Wade, 1987; Deschamps & Brown, 1983) suggests that when distinct task role assignments converge with category membership, threat to group identity and consequent intergroup bias will be reduced. To clarify these conflicting predictions, we created a cooperative contact setting in which members of two experimentally created groups worked together as a team. During the team membership phase, work roles either converged with or cross-cut initial group membership. Experiment 1 supported the predictions of the social categorization model; subjects in the cross-cut condition perceived greater similarity among team members and, in turn, showed less intergroup bias in reward allocation, than did those in the convergent condition. Two variables hypothesized to account for this outcome were the absence of negative task attitudes and the opportunity for personalization. In Expt 2, procedures likely to increase negative task attitudes and reduce personalization were inserted into the procedure of both conditions in an experimental design that paralleled that of Expt 1. Under these conditions, the differential effects of role assignment were eliminated. Taken together, these studies suggest that cross-cutting role assignment will reduce intergroup bias when it is implemented in a manner that does not arouse negative task attitudes but does provide opportunity for personalization of team mates.