We consider the covariance structure of health. Agents report their health status on the basis of a latent health stock that is determined by permanent and transitory shocks, and time invariant fixed effects. At age 25, permanent shocks account for 5% to 10% of the variation in health. At age 60, this percentage rise to between 60% and 80%. We document a gradient in which permanent shocks matter less for college-educated people and for women.