We analyzed the first 5 years of surveillance data from the Established Population for the Epidemiologic Studies of the Elderly (EPESE) in the Piedmont of North Carolina (n = 4162) to estimate the effect of a global self-rating of health on survival. Covariates used in Cox proportional hazard models included sociodemographic factors, chronic medical conditions, activities of daily living, use of health services, health risk behaviors, cognitive function, affective mood, and negative life events. Adjusted risk of mortality associated with poor (compared to excellent) self-ratings of health was significantly elevated among urban men only. Confounders of the association between survival and overall health assessment varied widely by subgroup. Pooled estimates from heterogeneous populations may mask significant subgroup differences both in the pattern of variables that mediate crude risk and also in the magnitude of residual risk of global self-ratings of health.