Within social psychology, it is well accepted that trait inference is the dominant tool for understanding others’ behavior. Outside of social psychology, a different consensus has emerged, namely, that people predominantly explain behavior in terms of mental states. Both positions are based on limited evidence. The trait literature focuses on trait ascriptions to persons, not explanations of behavior. The mental state literature focuses on explanations of ordinary behaviors (for which social scripts provide mental states), not of expectancy-violating behaviors. We examined the critical test case for the two opposing positions: explanations of expectancy-violating behaviors. Participants provided open-ended explanations of puzzling actions, which were content-analyzed for use of mental states, traits, and other causal background factors. Across four studies, three stimulus sets, and two subpopulations, people overwhelmingly offered mental states when explaining puzzling actions (compared with ordinary actions), while they struggled to generate traits and other background factors.