Different models of uncertainty aversion imply strikingly different economic behavior. The key to understanding these differences lies in the dichotomy between first-order and second-order ambiguity aversion, which I define here. My definition and its characterization are independent of specific representations of decisions under uncertainty. I show that with second-order ambiguity aversion, a positive exposure to ambiguity is optimal if and only if there is a subjective belief such that the act’s expected outcome is positive. With first-order ambiguity aversion, zero exposure to ambiguity can be optimal. Examples in finance, insurance, and contracting demonstrate the economic relevance of this dichotomy. This paper was accepted by Han Bleichrodt, decision analysis.