While some investigators believe that the concept of depression is a continuum with mild and severe forms reflecting essentially the same entity, most suspect that the concept is instead heterogeneous and consists of a group of discrete subtypes. If this is so, identifying subtypes is a major priority. Ultimately such subtypes must be understood in terms of their underlying neural and even molecular mechanisms. Yet in order to search for such mechanisms, we still must begin with clinical phenomenology. Two major subtypes of serious depressions have been proposed. Endogenous or melancholic depression is one, while bipolar depression is another. Thinking about both these subtypes tends to assume an underlying biogenic mechanism that is relatively autonomous, although not necessarily free of environmental influences. This paper examines a series of attempts to identify discrete subtypes of depression. One approach, used in a series of investigations, involves the use of mathematical techniques such as cluster analysis in order to identify phenomenologically similar subgroups within the depressive spectrum. This approach has consistently identified a melancholic or endogenous syndrome. Our attempts to validate the concept of endogenous depression through examining external correlates, such as family history, have been less successful. An alternate method for subtyping depression stresses that the bipolar subtype represents a discrete form of severe endogenously caused depression. We have examined the phenomenology of bipolar versus unipolar depression and found it to differ significantly in a number of respects. Thus, endogenous depression and bipolar depression may represent different phenomena.