We present a theory of choice in which attention to the features of options is determined by the decision maker’s categorization of the current problem in a set of problems she solved in the past. Categorization depends on goal-relevant and contextual problem-level features. The model yields heterogeneity in attention and choice in a given problem based on different past experiences, choice rigidity when categorization does not change despite new data, and choice instability when changes in irrelevant context cause re-categorization. We show that heterogeneous and unstable mental representations unify major biases in judgments and decision making.