We use data from a television game show, involving elementary lotteries and substantial prize money, as a natural experiment to measure risk attitudes. We find robust evidence of substantial risk aversion. As an extension, we esimate the various models using transformations of the ?true? probabilities to decision weights. The estimated degree of risk aversion increases further, while players tend to overestimate substantially their chances of winning. Constant Relative Risk Aversion (CRRA) and Constant Absolute Risk Aversion (CARA) utility specifications perform approximately equally well, with CARA having the advantage that the players? decisions do not depend on their initial wealth.