This study used the concept of diagnostic stability to examine the validity of three subtypes of major depression.Patients with major depressive disorder (N = 424) were assigned baseline diagnoses according to structured interviews and the Research Diagnostic Criteria. Follow-up evaluations took place at 6-month intervals over the next 5 years and annually for an additional 3 years. During this period 424, 246, 163, and 96 of the patients who had recovered from the index episode had one, two, three, and four recurrences, respectively, of major depressive disorder. The kappa statistic was used to quantify the likelihood that patients with the psychotic, agitated/retarded, or endogenous subtype of depression in a given episode would again manifest that subtype in subsequent episodes.The psychotic subtype showed the most enduring diagnostic stability across multiple subsequent episodes. Even after three intervening episodes, patients with baseline psychotic major depression were five times more likely to develop a psychotic depression than were other depressed patients. For all three subtypes, diagnostic stability was greater for contiguous episodes than for noncontiguous episodes. Psychotic, agitated/retarded, and endogenous subtypes showed significant stability after control for the bipolar/unipolar and primary/secondary distinctions. The endogenous subtype was stable among patients with primary depression but not among those with secondary depression.The psychotic subtype was the most valid of the subtypes tested from the perspective of diagnostic stability. The fact that stability across adjacent episodes exceeded stability across more distantly spaced episodes may reflect state-dependent determinants, and these are likely to vary by subtype.