Currency carry trade returns are on average large and non-normally distributed. While the literature has found different explanations for the existence of carry trade returns, the higher order moments of their return distribution still pose a puzzle. We propose a new model to explain these non-normal properties of currency carry trade returns, by assuming that agents are loss averse and overweight states with low probabilities. This combination of loss aversion and probability weighting is called crash-o-phobia. Using non-linear least squares and risk-neutral state prices implied by currency options, we estimate this crash-o-phobia model to price developed and emerging market currencies. The parameter estimates reveal crash-o-phobic beliefs and preferences with significant differences across currencies. Compared to a model with rational beliefs and CRRA utility, our crash-o-phobia model performs significantly better at explaining the whole distribution of currency carry trade returns.