BACKGROUND: This study aims to examine diagnosis-specific sickness absence as a risk marker for all-cause mortality. METHODS: Prospective occupational cohort (the GAZEL study). Medically certified sickness absence spells >7 days for 15 diagnostic categories, 1990-1992, were examined in relation to all-cause mortality, January 1993-February 2007. The reference group for each diagnostic category was participants with no spell >7 days for that diagnosis. The participants were French public utility workers (5271 women and 13 964 men) aged 37-51 years in 1990, forming the GAZEL study. Over the follow-up period, there were 144 deaths in women and 758 in men. RESULTS: 7875 employees (41.0%) had at least one spell of sickness absence >7 days over the 3-year period. The commonest diagnoses were mental disorders, musculoskeletal diseases, respiratory diseases and external causes in both sexes; genitourinary diseases in women, and digestive and circulatory diseases in men. Of these common diagnoses, mental disorders in women, hazard ratio (95% confidence intervals) 1.24 (1.1 to 1.4), and mental disorders 1.35 (1.3 to 1.5), digestive diseases 1.29 (1.1 to 1.6) and circulatory diseases 1.35 (1.2 to 1.6) in men were associated with mortality after adjustment for age, employment grade and sickness absence in all other diagnostic categories. CONCLUSIONS: Employees with medically certified absence spells of 1 week or more over a 3-year period had a 60% excess risk of early death. In women and men this excess risk was associated with some of the commonest diagnoses of sickness absence, in particular mental disorders. Sickness absence for mental disorders may be a useful early indicator of groups at increased risk of fatal disease.