The purpose of this study was to determine the diagnostic accuracy of medial temporal lobe atrophy (MTA) on MRI for distinguishing Alzheimer's disease from other dementias in autopsy confirmed cases, and to determine pathological correlates of MTA in Alzheimer's disease, dementia with Lewy bodies (DLB) and vascular cognitive impairment (VCI). We studied 46 individuals who had both antemortem MRI and an autopsy. Subjects were clinicopathologically classified as having Alzheimer's disease (n = 11), DLB (n = 23) or VCI (n = 12). MTA was rated visually using a standardized (Scheltens) scale blind to clinical or autopsy diagnosis. Neuropathological analysis included Braak staging as well as quantitative analysis of plaques, tangles and alpha-synuclein Lewy body-associated pathology in the hippocampus. Correlations between MTA and pathological measures were carried out using Spearman's rho, linear regression to assess the contributions of local pathologic changes to MTA. Receiver operator curve analysis was used to assess the diagnostic specificity of MTA for Alzheimer's disease among individuals with Alzheimer's disease, DLB and VCI. MTA was a highly accurate diagnostic marker for autopsy confirmed Alzheimer's disease (sensitivity of 91% and specificity of 94%) compared with DLB and VCI. Across the entire sample, correlations were observed between MTA and Braak stage (rho = 0.50, P < 0.001), per cent area of plaques in the hippocampus (rho = 0.37, P = 0.014) and per cent area of tangles in the hippocampus (rho = 0.49, P = 0.001). Linear regression showed Braak stage (P = 0.022) to be a significant predictor of MTA but not percent area of plaques (P = 0.375), percent area of tangles (P = 0.330) or percent area of Lewy bodies (P = 0.086). MTA on MRI had robust discriminatory power for distinguishing Alzheimer's disease from DLB and VCI in pathologically confirmed cases. Pathologically, it is more strongly related to tangle rather than plaque or Lewy body pathology in the temporal lobe. It may have utility as a means for stratifying samples in vivo on the basis of putative differences in pathology.