BACKGROUND AND PURPOSE-: The aim of the present study was to determine if outcome in terms of upper limb function at 6 months after stroke can be predicted in hospital stroke units using clinical parameters measured within 72 hours after stroke. In addition, the effect of the timing of assessment after stroke on the accuracy of prediction was investigated by measurements on days 5 and 9. METHODS-: Candidate determinants were measured in 188 stroke patients within 72 hours and at 5 and 9 days after stroke. Logistic regression analysis was used for model development to predict upper limb function at 6 months measured with the action research arm test (ARAT). RESULTS-: Patients with an upper limb motor deficit who exhibit some voluntary extension of the fingers and some abduction of the hemiplegic shoulder on day 2 have a probability of 0.98 to regain some dexterity at 6 months, whereas the probability was 0.25 for those without this voluntary motor activity. Sixty percent of patients with some early finger extension achieved full recovery at 6 months in terms of action research arm test score. Retesting the model on days 5 and 9 resulted in a gradual decline in probability from 0.25 to 0.14 for those without voluntary motor activity of shoulder abduction and finger extension, whereas the probability remained 0.98 for those with this motor activity. CONCLUSIONS-: Based on 2 simple bedside tests, finger extension and shoulder abduction, functional recovery of the hemiplegic arm at 6 months can be predicted early in a hospital stroke unit within 72 hours after stroke onset. © 2010 American Heart Association, Inc.