Misattributions people make about their own affective reactions can be used to measure attitudes implicitly. Combining the logic of projective tests with advances in priming research, the affect misattribution procedure (AMP) was sensitive to normatively favorable and unfavorable evaluations (Experiments 1-4), and the misattribution effect was strong at both fast and slow presentation rates (Experiments 3 and 4). Providing further evidence of validity, the AMP was strongly related to individual differences in self-reported political attitudes and voting intentions (Experiment 5). In the socially sensitive domain of racial attitudes, the AMP showed in-group bias for Black and White participants. AMP performance correlated with explicit racial attitudes, a relationship that was moderated by motivations to control prejudice (Experiment 6). Across studies, the task was unaffected by direct warnings to avoid bias. Advantages of the AMP include large effect sizes, high reliability, ease of use, and resistance to correction attempts.