OBJECTIVE: Development of a prognostic tool for patients with unresectable pancreatic cancer to distinguish between with low or high probabilities of survival 3 to 9 months after diagnosis. DESIGN: Data about individual patients from five studies were pooled. A multivariate proportional hazards model with time-dependent covariates was developed, including age, sex, and metastases. An extended model was developed on a subset of patients, including weight loss, pain, and jaundice at diagnosis. SETTING: Multicentre study, The Netherlands, Norway, USA, UK, and Canada. SUBJECTS: 1020 patients with unresectable pancreatic cancer. MAIN OUTCOME MEASURES: Prediction of prognosis. RESULTS: Patients with metastases, pain, or weight loss at diagnosis had a significantly poorer prognosis than the others. Older men had a worse prognosis than younger men, while older women had a better prognosis than younger ones. Patients with jaundice had a relatively good prognosis. Differences in survival among the studies were incorporated in a prognostic score chart. CONCLUSION: The prognostic score chart can be used to select patients with relatively low expectation of survival for endoscopic palliation, and patients with relatively high expectation for surgical palliation.