On the basis of a domain-specific theory of self-esteem, it was hypothesized that functionally distinct domains of self-esteem would predict aggression differentially. Participants completed self-report measures of self-perceived superiority, mate value, social inclusion, and global self-esteem, as well as of aggression. Self-assessed mate value emerged as a reliable, positive predictor, and social inclusion as a reliable inverse predictor, of self-reported hostility and aggression. In a subsequent laboratory experiment, in which participants had an opportunity to aggress against the source of positive or negative feedback about a personal essay that they had written, mate value again predicted increased aggression, whereas global self-esteem predicted decreased aggression. These main effects were moderated by the feedback manipulation, such that their respective simple effects were only present among participants that received negative feedback. Aggr. Behav. 00:1–11, 2005. © 2005 Wiley-Liss, Inc.