Differential diagnosis of patients whose course of illness includes substantial psychotic and mood syndromes is among the most challenging in psychiatry. The relative temporal preponderance of one or the other of these syndromes over course of illness forms the basis for distinctions among DSM-III-R diagnoses of schizoaffective disorder (SA), bipolar disorder (BPD), and schizophrenia (SZ); and such temporal assessments may be especially difficult to make reliably. Elsewhere we report relatively low reliability of SA and a tendency for it be “confused” with SZ and BPD. In this paper, we identify clinical variables that increase diagnostic differentiation. Data are from a Diagnostic Interview for Genetic Studies (DIGS) reliability study in which patients with independently assessed DSM-III-R lifetime diagnoses of SA-bipolar subtype,(SA-BP), BPD, and SZ were also clinically assessed and diagnosed by the DIGS on two occasions by two different interviewers blind to entry diagnoses. The relative strength of DIGS-based DSM-III-R diagnoses and individual DIGS clinical variables in predicting entry diagnoses is shown in a series of logistic regression analyses. Models incorporating DIGS variables are more predictive of entry diagnoses than models using DIGS diagnoses alone. Based on DIGS information, the SA-BP group is more clearly differentiated from the BPD group than from the SZ group. Different profiles of DIGS variables distinguish the groups. Findings are discussed in terms of their implications for nosologic research. Depression 3:309–315 (1995/1996). © 1996 Wiley-Liss, Inc .