Despite high-resolution preoperative imaging, many patients with apparently resectable pancreas cancer still undergo non-curative surgical procedures. We quantified factors potentially influencing surgical decision-making in order to generate an algorithm for optimizing the treatment of patients with pancreas cancer. We hypothesize that modern helical computed tomography (CT) and endoscopic ultrasound (EUS) can be used to predict the probability of a complete surgical resection (R0) as well as the likelihood of unsuspected metastases. METHODS: 180 patients underwent surgical exploration for pancreatic neoplasms between 2002 and 2005. All patients underwent modern helical CT using a pancreas mass protocol. EUS was obtained at the discretion of the treating physicians. In addition to preoperative clinical data and tumor markers, CT and EUS findings were compared with the operative results and pathology reports to quantify variables which may predict resectability and analyzed by a biostatistician using both univariate (chi-square & Wilcoxon rank-sum) and multivariate analyses. RESULTS: Of the 98 patients with pancreatic adenocarcinoma, 65 had margin-negative (R0) pancreatic resections, 12 had unsuspected metastases, 11 had positive margins, and 9 were locally unresectable. Resections included pancreaticoduodenectomy (64), distal pancreatectomy (10), and total pancreatectomy (3). 11 patients (14%) required portal vein (PV) resection. The addition of PV resection had no effect on the positive margin rate (p=0.52). By univariate analysis, predictors of an R0 resection included the absence of vascular involvement (p=0.02) and tumor size (p=0.04) by CT, and the EUS stage by AJCC criteria (p=0.02). Patients with suspected nodal disease by EUS were more likely to have metastases (36% vs. 13%, p=0.06). Excluding patients with metastases, vascular involvement by CT reduced the R0 resection rate from 80% to 45% (p=0.01). Moreover, each increase in the EUS stage portended a 2.7 fold increased probability of a non-curative procedure by multivariate logistic regression (p=0.004). The decrease in the R0 resection rate was particularly evident between EUS stages IIA and IIB, in which the rate of resectability dropped form 70% to 45% and the R0 rate fell from 55% to 36% (p=0.01). CONCLUSIONS: The assessment of vascular involvement by CT and preoperative stage by EUS were both highly predictive of resectability as well as R0 status. Using modern imaging, these data provide a probability algorithm for predicting which patients with pancreas cancer are likely to undergo a curative operation and form a basis for optimizing future treatment for this lethal disease.