The attraction effect occurs when the presence of an inferior option (the decoy) increases the attractiveness of the option that dominates it (the target). Despite its prominence in behavioral science, recent evidence points to the puzzling existence of the opposite phenomenon—a repulsion effect. In this paper, we formally develop and experimentally test a normative account of the repulsion effect. Our theory is based on the idea that the true values of options are uncertain and must be inferred from available information, which includes the properties of other options. A low-value decoy can signal that the target also has low value when both are believed to be generated by a similar process. We formalize this logic using a hierarchical Bayesian cognitive model that makes predictions about how the strength of the repulsion effect should vary with statistical properties of the decision problem. This theory may help account for several documented phenomena linked to the repulsion effect across both economic and perceptual decision making, as well as new experimental data. Our results shed light on the key drivers of context-dependent judgment across multiple domains and sharpen our understanding of when decoys can be detrimental.