Objective Marriage is considered to protect health via multiple mechanisms, but this effect may have weakened as marriage has become deinstitutionalized in the United States. This article tests for cross‐cohort decline in the protective effect of marriage. Methods Change in the association between marital status and subjective general health over three birth cohorts was estimated using the 1984–2011 Panel Study of Income Dynamics (N = 12,373). Analyses included least‐squares, random‐effects, and fixed‐effects regression models, representing increasingly conservative approaches to ruling out selection bias. Results Despite associations between marriage and better health among both men and women, estimated by least‐squares and random‐effects regression, the fixed‐effects models found health improvement relative to remaining unmarried only in very long (≥10 year) marriages, and only among women. This effect was completely attenuated among women in the youngest birth cohort. Conclusion The modest benefit of marriage for women's subjective health has eroded in recent cohorts.