The role of intracranial atherosclerosis in Alzheimer's disease (AD) has been a subject of debate since the first decade of the last century. The initial "vascular hypothesis" of AD was rejected after a series of mid-twentieth century gross anatomical postmortem studies that showed an inconstant relationship between intracranial atherosclerosis and senile dementia. These early studies did not utilize statistical methods, however, and the investigators did not appear to consider the possibility that intracranial atherosclerosis might have a probabilistic, rather than an absolute, effect on AD risk. Recent studies by three independent groups have found a significant statistical association between postmortem measures of circle of Willis atherosclerosis and AD. The present study was undertaken to further address the validity of this association in a large autopsy series, including cases diagnosed neuropathologically with vascular dementia (VaD) and non-AD dementias. Postmortem gross anatomical grading of circle of Willis atherosclerosis was performed in 397 subjects classified by neuropathological diagnosis, including 92 non-demented elderly controls, 215 with AD, 30 with VaD and 60 with non-AD dementias. Circle of Willis atherosclerosis was more severe in subjects with AD and VaD than in control subjects, while it was equivalent between control subjects and subjects with non-AD dementias. Increasing atherosclerotic grade increased the odds ratios (OR) for the diagnoses of both AD and VaD and also increased the ORs for both increased neuritic plaque density and higher Braak neurofibrillary tangle stage. The significance of these associations was retained after consideration of the effects of age, gender and the apolipoprotein E-epsilon4 allele. The results suggest that the statistical association between intracranial atherosclerosis and AD is not an artifact of diagnostic misclassification or of unequal distribution of the apolipoprotein E-epsilon4 allele.