We use the Panel Study of Income Dynamics (PSID) and the Medical Expenditure Panel Survey (MEPS) to study the relationship between marriage and health for working-age (20 to 64) individuals. In both data sets married agents are healthier than unmarried ones, and the health gap between married and unmarried agents widens by age. After controlling for observables, a gap of about 12 percentage points in self-reported health persists for ages 55-59. We estimate the marriage health gap non-parametrically as a function of age. If we allow for unobserved heterogeneity in innate permanent health, potentially correlated with timing and likelihood of marriage, we find that the effect of marriage on health disappears at younger (20-39) ages, while about 6 percentage points difference between married and unmarried individuals, about half of the total gap, remains at older (55-59) ages. These results indicate that association between marriage and health is mainly driven by selection into marriage at younger ages, while there might be a protective effect of marriage at older ages. We analyze how selection and protective effects of marriage show up in the data.