We use data from 298 showings of the television program "Affari Tuoi," which involves contestants making decisions between risky prospects with possible prizes of up to half a million euros, to estimate three models of decision-making under risk: Expected Utility, Rank-Dependent Expected Utility and Regret-Rejoice. We find that Regret-Rejoice does not significantly improve upon Expected Utility, while Rank-Dependent outperforms it. Interestingly, we find that the CARA specification fits significantly better than the conventionally-adopted CRRA specification. Crucially, we find a significant role for unobserved heterogeneity, implying that our estimates provide more superior estimates of risk attitude and of probability weighting than other studies.