PURPOSE: Information on patient ethnicity in hospital admissions databases is often used in epidemiologic and health services research. However, the extent of consistency of these data with self-reported ethnicity is not well studied, particularly for specific Asian subgroups. We examined agreement between ethnicity in records of a sample of members of five Northern California Kaiser Permanente medical centers with self-reported ethnicity. METHODS: Subjects were 3168 cases and 2413 controls aged 45 years and older from a study of fractures. Ethnicity recorded in the Kaiser admissions database (primarily inpatient) was compared with self-reported ethnicity from the study interviews. RESULTS: Among study subjects with available Kaiser ethnicity, sensitivities and positive predictive values of the Kaiser classification were high among blacks (0.95 for both measures) and whites (0.98 and 0.94, respectively), slightly lower among Asians (0.88 and 0.95, respectively), and considerably lower among Hispanics (0.55 and 0.81, respectively) and American Indians (0.47 and 0.50, respectively). Among Asian subgroups, the proportion classified as Asian was high among Chinese (0.94) and Japanese (0.99) but lower among Filipinos (0.79) and other Asians (0.74). Among the 228 (4%) subjects who self-identified with multiple ethnicities, 13 of 18 white + Hispanic subjects were classified as being white, and of the 77 subjects identifying as part American Indian, only one was classified as being American Indian in the Kaiser database. CONCLUSIONS: Given the importance of ethnicity information, medical facilities should be encouraged to adopt policies toward collecting high quality data.