History of psychiatric illness in 982 first degree relatives of 155 inpatients with major depression was examined to determine its clinical value for distinguishing melancholic from nonmelancholic major depressions. A positive family history of unipolar depression was not more common in the melancholic patients, although relatives of melancholic probands were more likely to have received electroconvulsive therapy or experienced nonaffective psychosis. Whereas the morbid risk estimate for unipolar depression was higher in the relatives of nonmelancholic probands, this appeared to be related to the manner in which morbid risk was calculated. Two methodological problems, variability in the risk periods of depressive subtypes, and the reliability of subtyping depressed family members, are discussed with respect to their implications for future family studies.