Simulation is a commonly used training tool in non-medical industries, but efficacy remains unproven in most health applications. Our objective was to evaluate satisfaction, knowledge, and self-confidence outcomes from a simulated diabetes training intervention consisting of: (a) 18 complex diabetes cases, (b) a web-based interactive interface, (c) a physiologic model to simulate outcomes of provider actions across a series of patient encounters, and (d) a library of feedback messages to critique and guide provider actions. 341 consented primary care residents in 19 residency programs were randomized to receive (n=177) or not receive (n=164) the intervention. Following the intervention, a satisfaction survey was completed by 94 (53%) intervention subjects, and results analyzed using mixed quantitative and qualitative methods. A knowledge and confidence questionnaire was completed by 220 (65%) of subjects and mean (95%CI) scores were compared by group, adjusting for program clustering. Responses were favorably higher than neutral for general satisfaction (93%), recommending to colleagues (91%), training adequacy (90%), navigation ease (95%), blood sugar displays (86%), drug & help links (76%), goal graphs (49%), and feedback (81%). The most valuable learning pertained to insulin use, general management, and goal achievement. The mean knowledge score (95% CI) was 5.31 (4.87-5.75) for intervention and 4.1 (3.69-4.50) for control subjects (p<.001). Self-confidence measures were higher for intervention compared to control subjects for: use of all drug classes (3.64 vs 3.09, p<.001), insulin use (4.12 vs. 3.36, p<.001), blood sugar interpretation (4.21 vs. 3.58, p< .001), individualizing goals (4.06 vs. 3.42, p< .001), and overall confidence (3.97 vs. 3.28, p< .001). In conclusion, the simulated learning program improved knowledge and confidence for diabetes management in primary care residents. Satisfaction was high, indicating good potential for broader dissemination.