Several recent studies have shown that when black and white students work together on biracial learning teams, they make more cross-racial friendships than students in traditional classrooms. This study is a secondary analysis of one of these studies, conducted to discover for whom the effects of cooperative learning are the greatest. The sample was 402 seventh and eighth grade students in twelve inner-city English classrooms. Classes were randomly assigned to team learning or control classes for a ten week period. The team learning classes used Student Teams-Achievement Divisions, a cooperative learning technique described by R.E. Slavin. Results indicated that the treatment increased cross-race friendships equally for students of different sexes, races and achievement levels. Further analyses revealed that new cross-race friends tended to be mutual instead of unreciprocated friends and to be among the first six friends named on the sociometric questionnaire rather than more distant choices. (Author/MK)