Although accurate prediction of survival is essential for palliative care, few clinical methods of determining how long a patient is likely to live have been established. To develop a validated scoring system for survival prediction, a retrospective cohort study was performed with a training-testing procedure on two independent series of terminally ill cancer patients. Performance status (PS) and clinical symptoms were assessed prospectively. In the training set (355 assessments on 150 patients) the Palliative Prognostic Index (PPI) was defined by PS, oral intake, edema, dyspnea at rest, and delirium. In the testing sample (233 assessments on 95 patients) the predictive values of this scoring system were examined. In the testing set, patients were classified into three groups: group A (PPI< or =2.0), group B (2.0<PPI< or =4.0), and group C (PPI>4.0). Group B survived significantly longer than group C, and group A survived significantly longer than either of the others. Also, when a PPI of more than 6 was adopted as a cut-off point, 3 weeks' survival was predicted with a sensitivity of 80% and a specificity of 85%. When a PPI of more than 4 was used as a cutoff point, 6 weeks' survival was predicted with a sensitivity of 80% and a specificity of 77%. In conclusion, whether patients live longer than 3 or 6 weeks can be acceptably predicted by PPI.