This article develops a rational analysis of an important class of apparent preference reversals-joint-separate reversals traditionally explained by the evaluability hypothesis. The "options-as-information" model considers a hypothetical rational actor with limited knowledge about the market distribution of a stimulus attribute. The actor's evaluations are formed via a 2-stage process-an inferential stage in which beliefs are updated on the basis of the sample of options received, followed by an assessment stage in which options are evaluated in light of these updated beliefs. This process generates joint-separate reversals in standard experimental designs. The normative model explains why the evaluability hypothesis works when it does, identifies boundary conditions for the hypothesis, and clarifies some common misconceptions about these effects. In particular, it implies that joint-separate reversals are not irrational; in fact, they are not preference reversals. However, in expanded designs where more than 2 options are jointly evaluated, the model predicts that genuine (and rational) preference reversals will sometimes emerge. Results of 3 experiments suggest an excellent fit between the rational actor model and the judgments of human actors in joint-separate experiments.