In an effort to identify features indicative of underlying bipolarity within the unipolar relatives of bipolar probands, we compared unipolar relatives of bipolars with unipolar relatives of controls. Using data from the Yale-NIMH Collaborative Study of Depression, we compared a number of demographic and clinical features individually, and then developed a logistic regression model for the differences found. Unipolar relatives of bipolars were generally similar to relatives of controls, but they were older and more likely to suffer from more severe, even psychotic, depression, and somewhat less likely to report a brief transition into their illness. A multiple logistic regression model for observed differences was highly statistically significant, but had limited ability to discriminate effectively between the two groups. These findings suggest that more stringent diagnostic criteria might be beneficial if unipolar relatives are counted as affected in linkage studies of bipolar disorder. The ability of this strategy to improve the "clinical phenotype" is limited, however, and other approaches may be needed to identify features of underlying bipolarity and thus to define "caseness" for unipolar relatives in linkage analyses of bipolar disorder.