Elderly persons (N = 310) attending outpatient psychiatric clinics were given an interview on the 30-item Geriatric Depression Scale (T. L. Brink et al., 1982; J. A. Yesavage et al., 1983) and received an independent psychiatric evaluation. A 3-step binary logistic regression showed that 2 items measuring positive affect and 2 others measuring negative affect combined to produce diagnostic performance comparable with the full scale. This result was cross-validated on a different sample (N = 134). A hierarchical measurement model in which the 4 items tapped 2 first-order factors (positive and negative affect), which in turn tapped a higher order construct of depression, provided excellent fit to the data.