Background/Aims: Members of minority groups are more likely than non-minorities to have worse health outcomes. Electronic medical records (EMR) data are often used for research on health disparities, and can help identify patient risk factors, but accuracy of ethnicity and race data in the EMR is often questioned. Aims: To compare HealthPartners Medical Group (HPMG) EMR data on ethnicity/race to self-reported classification. Methods: We compared percent agreement of self-reported and EMR data on ethnicity/race among 1719 patients who expressed interest in a hypertension clinical trial, completed an eligibility phone screen, and consented to EMR review. Patients classified their ethnicity/race in two questions: (1) Are you of Latino or Hispanic descent? (Yes, No, Don’t know, Refused); (2) How would you describe your race (Check all that apply) (American Indian/Alaskan Native, Asian, Black or African American, Native Hawaiian or Pacific Islander, White or Caucasian, Other, Don’t know, Refused). EMR race categories allowed multiple selections among: White, Black, Asian, Hispanic, American Indian, Other, and No Answer. Results: Race data was missing in 0.7% of phone screens and 1.2% of EMRs. Self-reported race was 79% white, 14% African American, 3% mixed, 2% other, 1% Asian, <1% Native American. In the EMR, race was recorded as 79% white, 14% African American, 3% mixed race, 1% Asian, <1% each for American Indian, and other. Overall agreement for race was 91.1%. EMR data agreed with self-report for 97% of those reporting white, 94% of those reporting African American, 83% Asian, 50% Native American, 9% mixed race and 6% other. Hispanic ethnicity was self-reported by 1.3% and coded for 2.4% in the EMR. EMR confirmed Hispanic ethnicity for 51% self-reporting Hispanic ethnicity. Conclusions: EMR race/ethnicity data was quite complete, and agreement of data sources was high overall and among whites, African Americans and Asians. Lower agreement was observed among Hispanic, Native American, and mixed race patients, subgroups that may be disproportionately affected by disparities. Some caution is needed in interpreting EMR-based results for those subgroups, and their proper classification in the EMR is warranted. Similar comparisons should be conducted in other settings and patient populations.