We report an experiment that infers true overconfidence in relative ability through actions, as opposed to reported beliefs. Subjects choose how to invest earnings from a skill task when the returns depend solely upon risk, or both risk and relative placement, enabling joint estimation of individual risk preferences and implied subjective beliefs of placing in the top half. We find evidence of aggregate overconfidence only in a treatment that receives minimal feedback on performance in a trial task. In treatments that receive more detailed feedback, aggregate overconfidence is not observed although identifiable segments of overand underconfident individuals persist.