BACKGROUND: The surgical management of duodenal gastrointestinal stromal tumors (DGIST) is poorly characterized. Limited resection may be technically feasible and oncologically safe, but anatomic considerations may compromise the resection margins due to the proximity of critical structures, thereby necessitating more extensive resections such as pancreaticoduodenectomy. METHODS: Patients undergoing surgery for DGIST at two institutions from 1994 to 2014 were identified. Clinicopathologic and survival data were analyzed to compare outcomes in patients treated with limited or radical resection. RESULTS: Sixty patients underwent surgery for DGIST. Pancreaticoduodenectomy was performed in 38 % while the rest underwent limited resections. The most common type of limited resection was wedge resection and primary closure (49 %) followed by segmental resection with an end-to-end or side-to-side duodenojejunostomy (27 %). The pancreaticoduodenectomy group tended to have larger tumors with the majority located in D2/3 (87 %) and at the mesenteric border (91 %). The pancreaticoduodenectomy group also had significantly greater intraoperative blood loss, longer operative time, longer hospital stay, and higher 90-day morbidity and readmission rates. The 5-year relapse-free survival, recurrence-free survival, and overall survival for the pancreaticoduodenectomy versus limited resection were 81 versus 56 % (p = 0.05), 64 versus 53 % (p = 0.5), and 76 versus 72 % (p = 0.6), respectively. A surgical algorithm based on the location and size of the tumor is proposed. CONCLUSIONS: Limited resection of DGIST is safe, but may be associated with lower 5-year relapse-free survival. Pancreaticoduodenectomy is recommended for selected patients with DGIST when an R0 resection cannot be performed without removing the ampulla or part of the pancreas.