Self-evaluations of health status have been shown to predict mortality, above and beyond the contribution to prediction made by indices based on the presence of health problems, physical disability, and biological or life-style risk factors. Several possible reasons for this association are discussed: (a) methodological shortcomings of previous studies render the association spurious; (b) other psychosocial influences on mortality are involved and explain the association; and (c) self-evaluations of health status have a direct and independent effect of their own. Four-year follow-up mortality data from the Yale Health and Aging Project (N = 2812) are used to explore these possibilities. The analysis controls for the contribution of numerous indicators of health problems, disability and risk factors, and also makes adjustments of standard errors for the complex sample design. The findings favor the third possibility, an independent effect, to the extent that the particular set of psychosocial factors examined did not explain the basic association, and to the extent that the control variables were an adequately comprehensive set.