Education is negatively associated with mortality for most major causes of death. The literature ignores that cause-specific hazard rates are interdependent and that education and mortality both depend on cognitive ability. We analyze the education-mortality gradient at ages 18-63 using Swedish register data. We focus on months lost due to a specific cause of death which solves the interdependence problem, and use a structural model that derives cognitive ability from military conscription IQ scores. We derive the educational gains in months lost and the selection effects for each cause of death, and quantify the selection contribution of observed characteristics and unobserved cognitive ability. In a standard Cox model that controls for observed IQ, primary education was associated with 6 months lost when compared to secondary education. In a structural model that accounts for cognitive ability the difference was 43% larger. In addition, the largest educational gains were achieved for the lowest education group in the reduction of external cause mortality. The educational gains in cardiovascular mortality was small, mainly due to large selection effects. These results suggest that educational differences in cause specific mortality may be biased by conventional Cox regression analyses.