Humans exhibit certain systematic context-dependent preference reversals when choosing among options that vary along multiple attribute dimensions. For instance, the attraction, similarity, and compromise effects each involves a change in relative preference between two options when a third option is introduced. Previously, such effects have been attributed to irrationality or sub-optimality in decision-making, or to specific architectural or dynamical constraints on cognition. We use a Bayesian model of multi-attribute choice to demonstrate that these effects naturally arise from three basic assumptions: (1) humans assess options relative to “fair market value” as inferred from prior experience and available options; (2) attributes are imperfectly substitutable, and scarce attributes are relatively more valuable; (3) uncertainty about market conditions and option values contributes to stochasticity in choice behavior. This work provides both a novel normative explanation for contextual modulation of choice behavior, and a means to predict choice as a function of past experiences and novel contexts.