Recently several papers have appeared which apply the techniques of multivariate analysis in attempts to show that patients suffering from depression can be divided into two groups, often called “endogenous” and “neurotic”, which tend to have different groups of symptoms and different outcomes with E.C.T. or other treatment. The purpose of the present paper is to look at this problem from the point of view of a statistician in an attempt to clarify the logic involved in the various procedures.