This work identifies how cross-group friendships are conceptualized and measured in intergroup research, investigates which operationalizations yield the strongest effects on intergroup attitudes, explores potential moderators, and discusses the theoretical importance of the findings. Prior meta-analyses have provided initial evidence that cross-group friendships are especially powerful forms of intergroup contact. Although studies of cross-group friendship have grown considerably in recent years, varied assessments leave us without a clear understanding of how different operationalizations affect relationships between friendship and attitudes. With a greatly expanded database of relevant studies, the authors compared friendship-attitude associations across a wide range of specific conceptualizations. Time spent and self-disclosure with outgroup friends yielded significantly greater associations with attitudes than other friendship measures, suggesting that attitudes are most likely to improve when cross-group friendships involve behavioral engagement. Processes underlying cross-group friendships are discussed, as are implications for future research and application.