The fit of the structure of DSM-III major depressive disorder to data from two large epidemiological surveys is assessed by latent class analysis. The surveys were conducted at the Baltimore and Raleigh-Durham sites of the National Institute of Mental Health (NIMH) Epidemiologic Catchment Area Program. Three classes are required to fit the data, and the third class bears a strong resemblance to major depressive disorder, although it requires slightly more symptoms to be present than DSM-III. The derived structure replicates successfully for Baltimore and Raleigh-Durham, with a prevalence of the major depression category of 0.9% for both sites.