The purpose of this study was to evaluate a simulated training program on resident physician ability to manage patients with diabetes mellitus (DM). 341 consented primary care residents from 19 residency programs were trained to use a web-based simulated diabetes management interface. They were then randomized by program to receive (n=177) or not receive (n=164) an intervention using 18 simulated learning cases with DM of varying severity and complexity (average time 15 minutes per case). Providers treated patients through an interactive electronic health record interface and were challenged to use any desired number of encounters to bring patients to care goals for A1c, blood pressure (BP), and LDL within 6 months of simulated time. Physiologic modeling simulated outcomes of provider actions over any follow-up time interval. Between encounters, providers received feedback to critique and guide actions. After the intervention period, control and intervention subjects were assigned 4 simulated assessment cases involving type 1 and 2 DM. Generalized linear mixed models were used to test for study arm differences in the proportion of residents bringing each simulated case to composite goal for A1c, BP, and LDL. 232 residents completed at least one assessment case (intervention (I) n=97, control (C) n=135). Residents were 52% female, 53% white, mean age 30, 44% family medicine, 53% internal medicine, and in post graduate years 1 (34%), 2 (35%), 3 (28.5%), and 4 (3%). The proportion of residents bringing patients to composite goal was significantly higher in the intervention group for all cases: case 1 - I 74.5%, C 21.3%, p < .001; case 2 - I 95.8%, C 61.1%, p < .001; Case 3 - I 82.6%, C 33.9%, P < .001; case 4 - I 96.7%, C 83.4%, p = .015. These results, in addition to previous studies showing improved outcomes of actual patients of primary care providers who completed similar learning programs, provide strong evidence to support integration of simulated diabetes training into medical residency curriculums.