The present study explores the effect of crossing social categorizations upon subsequent intergroup discrimination. In the simple categorization conditions, subjects were divided into groups either on an explicitly random basis or on the basis of a very trivial similarity. In the crossed categorization condition, these two categorizations were criss-crossed. After performing a perceptual estimation task, subjects had to evaluate the performance of the different groups in this task. Subsequently they had to evaluate the groups on general characteristics less directly related to task performance. There was significant intergroup discrimination favouring the own group in the two simple categorization conditions, but this discrimination was strongly reduced in the crossed categorization condition. This was true for both kinds of evaluations. Subjects of a no categorization condition exhibited no self-favouritism. The theoretical implications of the data are discussed.