We analyze data from the Bank of Italy's most recent recruitment rounds, in an effort to explain why men consistently score better than women. We focus on the pre-screening stage of the hiring process, a multiple-choice test, where men acquire a preliminary advantage. After observing a higher incidence of questions left blank for women, and a negative correlation between the share of unanswered questions and the final score, we run an experiment on scoring formulas to check for implicit discrimination linked to risk aversion; no evidence of such discrimination is found. Based on a follow-up questionnaire, we also study the role of composition effects. Nearly 40 per cent of the gap in test scores depends on the quality of the candidates: male graduates appear to self-select into the applicant pool more frequently than females do. A further 34 per cent is explained by the fact that the same characteristics tend to produce different effects based on gender. The remaining 26 per cent remains unexplained.