Complexity of the problem of choosing among uncertain acts is a salient feature of many of the environments in which departures from expected utility theory are observed. I propose and axiomatize a model of choice under uncertainty in which the size of the partition with respect to which an act is measurable arises endogenously as a measure of subjective complexity. I derive a representation of incomplete Simple Bounds preferences in which acts that are complex from the perspective of the decision maker are bracketed by simple acts to which they are related by statewise dominance. The key axioms are motivated by a model of learning from limited data. I then consider choice behavior characterized by a "cautious completion" of Simple Bounds preferences, and discuss the relationship between this model and models of ambiguity aversion. I develop general comparative statics results, and explore applications to portfolio choice, contracting, and insurance choice.