Some studies suggest that the relationship between education and health strengthens with age (cumulative disadvantage hypothesis), while other studies find that it weakens (age-as-leveler hypothesis). This research addresses this inconsistency by differentiating individual-level changes in health from those occurring at the aggregate level due to selective mortality. Using retrospective and prospective data from a nationally representative sample of U.S. adults, I examine educational differences in age-specific rates of disease prevalence, incidence, and survival. At the aggregate level, I find that educational differences in disease prevalence are largest at mid-life and then decline. At the individual level, however, disease incidence and mortality increase with age at a greater rate for less-educated persons compared to the well educated. These findings suggest that the cumulative disadvantage hypothesis explains how education affects the health of individuals with increasing age, whereas the leveling hypothesis describes the aggregated by-product of these educational disparities in health decline. (Contains 2 figures and 4 tables.)