We present a new experimental investigation into the cause of preference reversal. Although economists and psychologists have suggested a variety of possible accounts, the existing data does not adequately discriminate between them. Our design has two key novelties: a “pure” ordinal payoff scheme and a probabilistic valuation task. Relative to previous studies, this design offers enhanced control for economic explanations and new tests of psychological hypotheses. We find strong and systematic preference reversals that are inconsistent with a broad range of previous hypotheses. We explore and evaluate the explanatory strategies that survive exposure to our data.