We construct a health measurement model which combines panel data on self-reported health with a rich set of health measures from administrative medical records. Our estimated health model allows us to predict health status for the population at large. We account both for unobserved heterogeneity and for the persistence in unobserved health shocks. To account for inconsistent reporting in self-reported health we propose a model using two measures of self-reported health: the level of self reported health and the self-reported health change. We show that using both measures substantially increases the estimated persistence in health status. We use predicted health status to study the evolution of health as individuals age. Moreover, we analyze how health interacts with economic variables and education. We find a strong gradient in education; the age at which health starts to decline at a greater rate differs by education and gender.