Reliability of self-reported age at onset of major depressive disorder was studied in a sample of 335 subjects who were ascertained from a large epidemiologic survey conducted in several U.S. communities and who were interviewed blindly at two different times. Reliability was generally good for these subjects who met DSM-III criteria for depression at two interviews. A large proportion of the variability in the difference of test-retest values can be accounted for by recency of last episode of depression and interactions of age with duration of illness, having been treated for a mental health problem, and comorbidity of other mental disorders. Interactions between duration and comorbidity and between geographic region and treatment were also significant. Contrary to previous studies which do not consider the interval between current age and age at onset and which suggest that reliability diminishes with age, our findings show that older respondents tend to systematically decrease and not increase age at onset across the two interviews. These findings do not support the hypothesis that recently reported secular changes in major depression, including a decreased age at onset and higher rates in younger as compared to older cohorts, can be explained by a differential reporting effect. Furthermore, the findings suggest that factors which contribute to variability in an individual's age at onset should be incorporated in genetic and clinical studies of major depression.