This paper experimentally tests whether violations of Savage’s (1954) subjective expected utility theory decrease if the ambiguity of an uncertain decision situation is reduced through statistical learning. Because our data does not show such a decrease, existing models which formalize ambiguity within an Anscombe-Aumann (1963) framework – thereby reducing to expected utility theory in the absence of ambiguity – are violated. In contrast, axiomatic models of prospect theory can accommodate our experimental findings because they allow for violations of von Neumann and Morgenstern’s (1947) independence axiom whenever uncertain decision situations transform into risky decision situations for which probabilities are known.