Sometimes evolutionary biologists explain the prevalence of a social trait by hypothesizing that there was group selection for the trait. But philosophers of biology (and biologists themselves) disagree about how to define group selection: each candidate definition has some counter-intuitive consequences, and each candidate definition strikes some of its opponents as intuitively "ad hoc". To make progress on this issue, Clarke argues, we need to move beyond this trading of intuitions. Instead, we should ask: what theoretical role are group selection hypotheses meant to play? Clarke suggests that the role of group selection hypotheses is to offer causal explanations that "avoid overlapping" with individual selection hypotheses---in a sense that he will make precise. Clarke then explores some consequences of this suggestion. Firstly, there are at least two distinct definitions of group selection that are theoretically motivated. Secondly, the definition that is most appropriate, to apply to a given biological scenario, will depend upon one's explanatory interests.