Sir Aubrey Lewis studied 61 depressives in considerable detail, principally cross-sectionally but also by reviewing progress. He concluded that he could find no qualitative distinctions between the depressed patients and thus established himself as a strong and influential advocate of the unitary view of depression (i.e. that depression varies dimensionally, not categorically). Subsequently, Kiloh & Garside (proponents of the binary view of two depressive ‘types’) coded the Lewis data and undertook a principal components analysis. They claimed success in distinguishing ‘endogenous’ and ‘neurotic’ depressive types within Lewis' sample. In this paper we re-analyse the data set using both a latent class categorical approach and mixture analyses. We suggest that any demonstration of sub-types was limited by relative homogeneity of the sample (in that up to 80% had probable or possible psychotic conditions), and by Lewis rating a number of important features (e.g. delusions) dimensionally rather than categorically. Nevertheless, we identify one categorical class (essentially an agitated psychotic depressive condition) and a residual (presumably heterogeneous) class. The presence of those two classes was supported by demonstrating bimodality in composite scores derived from the fourteen differentiating clinical features (and not evident when all clinical features were considered), and formally confirmed by mixture analyses. Membership of the categorical class was determined principally by psychotic features (delusions and hallucinations) and by objectively-judged psychomotor disturbance, and we consider the nature of that ‘class’. Lewis' data set is unusual (in having self-report and observationally rated data), and historically important in demonstrating that conclusions may depend on the choice of variables examined and analytical approaches.