We theoretically show that loss-averse agents are more likely to lie to avoid receiving a low payoff after a random draw, the lower the ex-ante probability of this bad outcome. The ex-ante expected payoff increases as the bad outcome becomes less likely, and hence the greater is the loss avoided by lying. We demonstrate robust support for this theory by reanalyzing the results from the extant literature and with two new experiments that vary the outcome probabilities and are run doubleanonymous to remove reputation effects. To measure lying, we develop an empirical method that estimates the full distribution of dishonesty