Background: The present study was designed to assess onset and persistence of late-life depression, systematically comparing the factors associated with prevalence, onset and prognosis. Methods: The data were derived from a large (n = 2200), random, age and sex stratified sample of the elderly (55-85 years) in The Netherlands. Using a 3-year, prospective longitudinal design, both the onset and the persistence of depression were assessed. Depression was measured using the Center for Epidemiologic Studies Depression Scale. Risk factors associated with prevalence, onset and persistence were compared using both bivariate and multivariate analyses. Results: In those not depressed at index assessment, the onset of depression was 9.7%. Among those depressed at baseline, persistence occurred in 50.4%. Risk factors predicting onset were almost identical to those associated with prevalence. Persistence was predicted by very few factors (external locus of control and chronic physical illness). Conclusions: The data suggest that cross-sectional studies are biased due to their overrepresenting chronic depressive episodes. However, the risk factors derived from cross-sectional studies do seem to adequately reflect factors associated with onset. The prognosis is not adequately predicted by variables usually included in epidemiological studies of late life depression. It is speculated that including more biological correlates of depression and data concerning positive life-changes may improve our understanding of the prognosis of late life depression. © 2001 Elsevier Science B.V.