OBJECTIVES: The aims of this study were to assess whether the site of lesion is predictive of upper limb recovery after stroke and to determine whether this information adds to the predictive ability of the clinical examination. MATERIAL AND METHODS: Forty-five patients were examined at entry to the study and at 2 and 12 months after stroke. The Brunnström-Fugl-Meyer test was used as outcome measurement. Predictor variables included clinical parameters and classifications of lesion site (obtained by CT/MRI). RESULTS: Correlation analysis revealed small to moderate relationships between lesions of subcortical structures and arm outcome at 2 months. In multiple regression analysis, the best model for predicting recovery at 2 months was found to be a combination of the clinical parameters with a purely subcortical lesion. Motor recovery at 12 months was best predicted by the clinical tests alone. The results further indicated that patients with subcortical damage tended to take longer to recover. CONCLUSIONS: Clinical assessment is most useful for determination of the prognosis of upper limb recovery after stroke. Neuroanatomical parameters measured by CT or MRI can only act as an adjunct.