A total of 162 patients referred for rehabilitation were followed-up for two years. Using the t test and the chi-square test, 15 variables distinguishing the 107 survivors from the 55 who died were identified. Multiple regression analysis identified three factors that gave a predictive equation which correctly identified 92% of the two-year survivors and 48% of those dying in that time. These factors were the severity of the stroke (represented by the patient's walking ability), the presence of pre-existing cardiovascular disease (indicated by a history of myocardial infarction), and the patient's age. Pre-stroke motivation or fitness may also influence long-term survival. The technique of using multiple regression analysis allows a clearer separation of the important prognostic factors than the more usual comparative statistical methods.