The possible modifying effect of social relations on the association between depression and mortality was examined in a community-based cohort study. A total of 3,777 randomly selected persons 65 years of age and older in southwest France were followed over a 5-year period from 1988 in the Personnes Agees Quid (PAQUID). At study entry, the prevalence of elevated depressive symptomatology was 12.9% for men and 14.7% for women, and the reported relative isolation was 14.1% for men and 26.0% for women. During a total of 16,984 person-years of follow-up, 849 deaths occurred. Among participants with high levels of depressive symptomatology, the age-adjusted mortality rate ratio was 2.10 (95% confidence interval 1.7-2.7) in men and 1.76 (95% confidence interval 1.4-2.3) in women. When compared with individuals with the most connections, men and women with few social network connections were also at increased risk of mortality: age-adjusted rate ratio = 2.69 (95% confidence interval 1.9-3.8) for men and 1.56 (95% confidence interval 1.0-2.4) for women. Satisfaction with social support had a small but nonsignificant effect on mortality risk. For women, the excess risks due to depressive symptoms and few network connections are observed only in the 65- to 74-year age group, after adjusting for health and health behaviors. Social relations did not significantly modify the depression-mortality associations for either men or women, although the depression-mortality effect was reduced by 12.8% in men. The latter findings do not appear to be compatible with the buffering hypothesis, whereby we would expect social relations to decrease the depression-mortality association. Nonetheless, there are independent effects from these two factors, and older men who are depressed and not socially connected are at increased risk of dying earlier.