OBJECTIVE: To determine whether a specific component of hostility (i.e. cognitive or behavioural) may predict suicide in a prospective design, controlling for depressive mood. METHOD: In 1993, 14,752 members of the 'GAZ et ELectricité' (GAZEL) cohort (10,819 men, mean age=49.0 years; 3933 women, mean age=46.2 years) completed the Center for Epidemiologic Studies Depression Scale and at least one subscale (i.e. cognitive or behavioural hostility) of the Buss and Durkee Hostility Inventory. Dates and causes of death were obtained annually. RESULTS: During a mean follow-up of 15.7 years, 28 participants completed suicide (24 men, four women). Suicide was predicted by depressive mood [relative index of inequality (RII) (95% CI)=8.16 (1.97-33.85)] and cognitive hostility [RII (95% CI)=10.76 (2.50-46.42)], but not behavioural hostility [RII (95% CI)=1.37 (0.38-4.97)]. These associations remained significant after adjustment for potential confounders. After mutual adjustment, however, suicide remained significantly associated with cognitive hostility [RII (95% CI)=8.87 (1.52-51.71)] (RII reduction: 34.6%), but no longer with depressive mood [RII (95% CI)=2.03 (0.41-10.07)] (RII reduction: 79.1%). CONCLUSION: Cognitive rather than behavioural hostility is associated with an increased risk of suicide, independently of baseline depressive mood.