Race and ethnicity are important predictors of health care access and outcomes, but quality of their documentation in the healthcare system is often problematic. To study the agreement between Medicare and Medicaid descriptions of race and ethnicity in older beneficiaries. Quasiexperimental design in a natural practice setting. New Jersey. 153,241 dually enrolled participants in Medicare and Medicaid. Agreement rates between administrative databases on recipients' race and ethnicity. Agreement between Medicare and Medicaid on the recipients' race and ethnicity was modest (kappa = .58; 95% CI, .57-.58) for men and women alike and across different age groups. Depending on whether Medicare or Medicaid was used as the reference standard, the relative agreement rates for race and ethnic group assignments varied. For example, using Medicare as the reference, the relative agreement rate was 84% for whites, 74% for blacks, 61% for others, 23% for Hispanics, and only 5% for Asians. Using Medicaid as the reference, a different pattern emerged. However, such gradients of agreement rates across racial groups were observed in both programs. Medicare and Medicaid reported different percentages of all race and ethnicity groups, with Medicaid reporting greater proportions of White and Black beneficiaries, and Medicare reporting greater proportions of Hispanic, Asian, and Other groups. Depiction of race and ethnicity data in large government health insurance programs is approximate at best and often contradictory from one program to another. This can impede efforts to study the relationship between these important characteristics and health care utilization and outcomes.