OBJECTIVES: To determine which symptoms of depression are most likely to increase the risk of mortality in a biracial sample of older adults. DESIGN: Cross-sectional and longitudinal study. SETTING: Baseline and mortality follow-up in urban and rural North Carolina. PARTICIPANTS: Four thousand one hundred sixty-two African-American and white elders aged 65 to 105 at baseline (mean age 73). MEASUREMENTS: Sociodemographic and health factors and four subscales of the Center for Epidemiologic Studies Depression Scale (negative affect, positive affect, somatic symptoms, and interpersonal function) were determined at baseline (1986-87). Mortality was assessed over 10 years of follow-up (through 1996). RESULTS: Fifty-one percent of the sample died over the 10-year follow-up. In controlled Cox proportional hazards modeling, those who scored lower on the positive affect scale were significantly more likely to die over the 10-year follow-up (hazard ratio=1.12, 95% confidence interval=1.05-1.18). For those who scored higher on the negative affect scale, the somatic scale, and the interpersonal scale, there was no increased risk for mortality in controlled analyses. CONCLUSION: These findings suggest that subjective views of well-being may be more important predictors of mortality in older adults than the classic symptoms of depression, such as negative affect and somatic symptoms.