AbstractWe match data on performance in a multiple‐choice examination with data on risk preferences from a classroom experiment. Students who are more loss averse leave more questions unanswered and perform worse in the exam when an incorrect answer is penalized compared with no answer. Thus, loss aversion parameters extracted from lottery choices in a controlled experiment have predictive power in a field environment of decision‐making under uncertainty. Furthermore, the degree of loss aversion appears to be persistent over time, as the experiment was conducted three months prior to the exam. Important differences across genders are partly explained by differences in loss aversion.