Summary The National Institute of Neurological and Communicative Disorders and Stroke/Alzheimer Disease and Related Disorders Association criteria for probable Alzheimer disease (AD) require exclusion of non-AD dementia-producing conditions but do not specify how the non-AD conditions are to be identified. We addressed this issue for the case of cerebrovascular disease (CVD) by defining exclusion rules based on commonly described clinical features: (a) history of strokelike episodes; (b) history of stepwise cognitive decline; (c) focal deficits on neurological examination; and (d) evidence of significant CVD on neuroimaging. We applied these rules retrospectively to clinical records for 92 cognitively impaired patients who otherwise met criteria for probable AD and whose brains were subsequently available for postmortem examination. We used Fisher's exact test to assess the effectiveness of the exclusion rules in predicting the presence of CVD on autopsy. Prediction was better than chance when all four clinical features were used together (p = 0.0008) and when the stepwise decline or neuroimaging criteria were used alone (p = 0.03 and p = 0.05, respectively). Overall, the CVD exclusion rules were deficient because of low accuracy (50.0%) and low sensitivity (52.6%). These results support provisional use of the CVD criteria chosen for this study but suggest that modifications are needed for acceptable diagnostic accuracy and sensitivity to be achieved.