The moderate utility model represents the probability of choosing an option in a pairwise comparison as an increasing function of utility difference divided by a dissimilarity metric. We provide a single, directly testable property that characterizes the model: choices are moderately transitive. We show that the model allows the analyst to accommodate well-known failures of strong transitivity, while retaining significantly more empirical bite than weak transitivity, achieving a useful compromise. Extending the analysis to the domain of risky choice, we introduce and characterize the moderate expected utility model, and we show how the analyst can measure utility and dissimilarity from observed choice behavior.