While the Mini-Mental State Examination (MMSE) was originally developed to screen for dementia and delirium, many neurologists use this measure as a screening instrument for 'cognitive impairment' in hospitalized stroke patients. However, the validity of the MMSE as such has never been evaluated in acute stroke. We administered the MMSE in addition to a neuropsychological examination covering six cognitive domains to 34 stroke patients (mean interval between stroke and examination, 6.5+/-2.9 days) and 34 healthy controls. The area under the receiver operating characteristic curve (AUC) was calculated in addition to the sensitivity and specificity for various cut-off points on the MMSE. Seventy percent of the patients were impaired in at least one cognitive domain. The accuracy of the MMSE in detecting cognitive impairment was no better than chance (AUC = 0.67; p = 0.13). No optimum MMSE cut-off value could be identified. The MMSE is particularly insensitive to impairments in abstract reasoning, executive functioning, and visual perception/construction.