Globally, more adolescents die from road traffic fatalities than from any other cause, and males are significantly more vulnerable than females. Driver education interventions directed at males are less likely to succeed than those directed at females, and stronger optimistic bias and overconfidence bias have been implicated as likely reasons. We report results from a quasi-experiment conducted in Serbia, targeting male and female adolescents. Stratified by size, forty schools were randomly assigned to either a personal-narrative intervention or a no-intervention control arm. Data were collected before the intervention (N = 1449) and again six months later (N = 1072). Risk perceptions improved for both males and females, and injunctive norms improved for females. Improvements in overconfidence bias and descriptive norms were predictive of improvements in high-risk driving behaviors. A significant interaction between improvements in injunctive norms and the intervention revealed that males whose injunctive norms improved were significantly more likely to be affected by the intervention, compared to the other groups. Implications for interventions are discussed.