Objectives– The study was aimed at improving the accuracy of prognosis for recovery of function in patients suffering a first stroke. Materials and methods– Two-hundred and forty-eight patients were enrolled. The mean interval since the stroke was 23 days. Patients entered a rehabilitation program lasting 60 days. The predictive value of 12 factors were analysed, namely motor, cognitive and sphincter subitems of Functional Independence Measure at admission (FIM-a), age, sex, education, body mass index (BMI), depression, neglect, aphasia, ideomotor and constructive apraxia. FIM score at discharge was the dependent variable. Results– A multiple regression revealed that only age, cognitive and sphincter subitems of FIM-a, neglect and ideomotor apraxia were significantly associated with outcome. Moreover, these factors accounted for only 72% of the variance in outcome scores. A decision of unfavourable prognosis on the basis of a FIM-a value lower than 40 was incorrect in 2.8% of the patients in this study and in 8.2% of those having a FIM score lower than 40. Conclusions– The use of statistical methods to examine the outcome after stroke is useful for expressing probability on a group basis but is unsuitable for determining the prognosis of individual patients. Such data should not be used for fiscal management. A significant minority of patients presenting with a FIM lower than 40 can regain a useful measure of independence. The errors in prognosis based upon available methods, although small, have unacceptable effects in human terms if they lead to the clinical decisions which deny patients rehabilitation. All of the patients should therefore be admitted for rehabilitation after their first stroke. Severe comorbidity requires special attention.