OBJECTIVE: To measure actual and perceived resident physician knowledge of underserved patient populations. PARTICIPANTS AND METHODS: Using the existing literature on vulnerable patient populations and interviews with experts in the field, we designed a cross-sectional, multi-institutional survey to assess actual and perceived resident knowledge of topics related to underserved populations. The survey of actual knowledge consisted of 30 multiple-choice questions, and the survey of perceived knowledge consisted of 15 items based on 3-point Likert scales of confidence. RESULTS: A total of 498 surveys were completed at 18 residency programs representing 7 different specialties at 10 US institutions. Assessment of perceived knowledge demonstrated that residents were very confident only 14.0% of the time, somewhat confident 66.4% of the time, and not at all confident 19.6% of the time. Assessment of actual knowledge revealed that the average percent correct across all 30 questions was 38%. Women scored better than men (average score, 40.6% vs 36.0%; P=.01), and African Americans scored higher than members of other racial or ethnic groups (average score, 43.5% vs 38.0%; P=.04). Associations between residents' perceived and actual knowledge were generally high. CONCLUSION: For the US residents surveyed, the actual and perceived knowledge about most topics relating to underserved populations was low, whereas associations between actual and perceived knowledge were high. These findings suggest the need to improve resident education regarding underserved patient populations.