Good friends can bolster health, happiness, and longevity. But makes for a good friend? There is a clear picture of this: According to both intuition and data, we become friends with people who are similar, familiar, and proximate to us, prefer friends who are kind and trustworthy, and disfavor friends who are vicious, exploitative, or indifferent. We propose, however, that this clear picture is partly a misrepresentation. Taking an evolutionary approach, we suggest that our friend preferences are target-specific. First, real-world behavior is target-specific (e.g., people are kind to some and vicious to others); friends’ behavior toward Person A does not necessarily predict their behavior toward Person B. Second, the fitness consequences of our friends’ behavior depend on targets—whether friends’ kindness or viciousness is directed toward us, our kin, our rivals. Further, evolutionary theories of friendship imply that we benefit when friends are preferentially prosocial toward us (versus indiscriminately prosocial). Across studies—some pre-registered, cross-cultural—we investigated how participants would prefer their ideal friends to behave toward them, strangers, and participants’ rivals. We predicted and found that (1) when friend preferences are unspecified (as in previous work), unspecified preferences track self-directed preferences and echo existing desires for prosociality, but people (2) also want friends to be preferentially prosocial toward them versus strangers (or rivals), and (3) prefer vicious (versus prosocial) friends—when friends behave toward rivals. Findings challenge intuitions and long-held conclusions about friend preferences and suggest a target-specific approach has utility for reexamining partner preferences.