Aims:To construct a predictive model for overall survival (OS) in unresectable pancreatic cancer (PaC) undergoing systemic chemotherapy and to confirm its accuracy in an independent cohort. Patients and methods: The training set (Ts) and the validation set (Vs) included 93 patients (median age=71 years) and 75 patients (median age=76 years). In the Ts, we examined variables linked to OS by uni-and multivariate analyses and constructed a predictive model for OS. Next, we evaluated the reproducibility of the proposed model in the Vs. Results: In the multivariate analysis for the Ts, PaC stage IV (P=0.0020) and carbohydrate antigen (CA) 19-9 >437.5 IU/l (P=0.0237) were identified to be significant factors associated with OS. Patients with PaC stage IV or not were given a score of 1 or 0, whereas patients with CA19-9 >437.5 IU/l or <437.5 IU/l were given a score of 1 or 0. Sum of the point of PaC stage (0 or 1) and CA19-9 (0 or 1) was defined as "PaC-CA score". In the Ts, there were 16 patients with score 0, 40 with score 1 and 37 with score 2, while in the Vs, there were 9 patients with score 0, 32 with score 1 and 34 with score 2. Overall P values reached significance in the Ts (P=0.0002), the Vs (P=0.0029) and the combined Ts and Vs (P<0.0001) among patients with PaC score 0, 1 and 2. Conclusion: PaC-CA score can be helpful for risk stratification in PaC patients undergoing systemic chemotherapy.