In 1982-1983 a random sample of 1486 people aged 65 years and above was generated from general practitioner lists; 1070 were interviewed in the community using the Geriatric Mental State and a Social History questionnaire. The cohort was followed up by interview 3 years later. At year 3 the diagnostic computer program AGECAT diagnosed 44 incident cases of depression. Information from the depressed group's initial and further interviews was compared with a control group (which excluded cases of affective or organic mental illness). Univariate analysis yielded three factors that were significantly associated with the development of depression 3 years later: a lack of satisfaction with life; feelings of loneliness; and smoking. Multivariate analysis confirmed their independent effects and revealed 2 further factors attaining significance: female gender and a trigger factor, bereavement of a close figure within 6 months of the third-year diagnosis. Some other factors traditionally associated with depression, such as poor housing, marital status and living alone, failed to attain significance as risk factors.