Dose escalation for pancreas cancer is limited by the tolerance of the duodenum, especially with stereotactic body radiotherapy (SBRT). This study reports on the dosimetric determinants of duodenal toxicity with single-fraction radiotherapy. Fifty-three patients with locally advanced unresectable pancreatic adenocarcinoma received 25 Gy in a single fraction. Common Terminology Criteria for Adverse Events v3 (CTCAE) was used to score all duodenal toxicity. Dose–volume histogram (DVH) endpoints evaluated include V5 (volume of duodenum in cm3 that received 5 Gy), V10, V15, V20, V25, and Dmax (maximum dose to 1 cm3). Normal tissue complication probability (NTCP) was evaluated with a Lyman model through estimation of the parameters m, n, and TD50 using the log-likelihood method. Univariate and multivariate analysis was conducted with Kaplan-Meier and Cox-regression models, respectively. Ten patients (19%) experienced Grade 2–4 duodenal toxicity. The median time to toxicity was 6.6 months (range, 1.6–10.5 months). The 6- and 12-month actuarial rates of duodenal toxicity were 12% and 32%, respectively. The V10 - V25, and Dmax all correlated significantly with duodenal toxicity (p < 0.05). In particular, V15 ≥13 cm3 and V15 <13 cm3 yielded 1-year duodenal toxicity rates of 64% and 6.2%, respectively, (p = 0.001); V20 ≥6.6 cm3, and V20 <6.6 cm3 gave toxicity rates of 61% and 4.8%, respectively, (p = 0.004); and Dmax ≥24.6 Gy and Dmax <24.6 Gy gave toxicity rates of 73% and 4.8%, respectively, (p = 0.001). Lyman NTCP model optimization generated the coefficients m = 0.15, n = 5.1, and TD50 = 22.5 Gy. Gender, age, tumor location, palliative bypass surgery, presence of a biliary stent, and chemotherapy did not correlate with toxicity on univariate analysis. Only the Lyman NTCP model remained significant in multivariate analysis (p = 0.028). Multiple DVH endpoints, as well as a Lyman NTCP model, are strongly predictive of duodenal toxicity. These dose constraints will be valuable in future studies with abdominal SBRT.