Diffusion MRI abnormality correlates with pathology in animal ischemic stroke models. A combined retrospective and prospective analysis of consecutive patients over a 3-year period who had a clinical diagnosis of probable new ischemic stroke, underwent diffusion MRI, and were later studied at autopsy was performed. Inclusion criteria for the retrospective analysis were 1) symptom onset within 14 days of presentation, 2) diffusion MRI within 28 days of symptom onset, and 3) autopsy within 16 weeks of symptom onset. Patients with suspected further infarcts between MRI and autopsy were excluded. The locations of all areas of MRI abnormality were identified by a blinded neuroradiologist, and recent infarcts were identified by review of pathologic records and microscopic slides. Eleven patients were identified who fulfilled inclusion criteria, with 25 discrete pathologic infarcts. Diffusion MRI abnormality corresponded to pathologically verified infarction in 23 cases, was present in two locations where no pathologic infarct was identified, and was absent in two locations where an infarct was present at autopsy. In two cases, despite clinical suspicion of acute ischemic stroke, no MRI abnormality or pathologic infarct was found. The sensitivity and specificity of diffusion MRI were 88.5% (95% CI, 69.9% to 97.6%) and 96.6% (95% CI, 91.5% to 99.1%). Accuracy was 95.1% (95% CI, 90.2% to 98%). Three further patients who died during the course of the retrospective analysis were studied prospectively, and are described separately. These findings suggest high accuracy of diffusion MRI for detection of ischemic infarction compared with pathologic examination.