OBJECTIVES: To examine the association between self assessed driving while sleepy and the risk of serious road traffic accidents (RTAs). DESIGN: Prospective cohort study. SETTING: France. PARTICIPANTS: 13 299 of the 19 894 living members of the GAZEL cohort, workers and recent retirees of a French national utility company followed up since 1989. MAIN OUTCOME MEASURES: Frequency of driving while sleepy in the previous 12 months, reported in 2001; rate ratios for serious RTAs in 2001-3, estimated by using generalised linear Poisson regression models with time dependent covariates. RESULTS: The risk of serious RTAs increased proportionally with the frequency of self reported driving while sleepy. After adjustment for sociodemographic characteristics, driving behaviour variables, work conditions, retirement, medical conditions and treatments, depressive symptoms, and sleep disorders, the adjusted rate ratios of serious RTAs for participants who reported driving while sleepy in the previous 12 months "a few times" or "once a month or more often" were 1.5 (95% confidence interval 1.2 to 2.0) and 2.9 (1.3 to 6.3) respectively compared with those who reported not driving while sleepy over the same period. These associations were not explained by any reported sleep disorders. CONCLUSIONS: Self assessed driving while sleepy was a powerful predictor of serious RTAs, suggesting that drivers' awareness of their sleepiness while driving is not sufficient to prevent them from having RTAs. Messages on prevention should therefore focus on convincing sleepy drivers to stop driving and sleep before resuming their journey.