Elected officials are increasingly extreme. Research trying to understand this trend has tended to focus on structural factors, such as primary elections and changes in the supply of candidates. Less emphasis has been placed on psychological perspectives. The current research advances such a perspective. Leveraging research on attitudes, we investigate when and why people prefer extreme over moderate candidates from their own party. We posit that the identity relevance of people’s attitudes plays a key role. Specifically, we propose that identity relevance fosters attitude extremity, which in turn promotes a preference for extreme over moderate candidates. Across six main studies (N = 3136) using a variety of political issues, operationalizations of identity relevance, instantiations of candidate extremity, and experimental paradigms (including two studies with human-LLM interactions), we find support for this hypothesis. Our findings suggest that as attitudes become more identity relevant, they become more extreme, leading individuals to prefer extreme over moderate candidates from their party. These results shed light on when and why people prefer extreme over moderate candidates, contribute to a nascent literature on the identity relevance of people’s attitudes, and advance our understanding of the antecedents and consequences of attitude extremity.