The medial temporal lobe of the brain is important for normal cognitive function, notably for memory, and is the region with the most extensive pathological change in Alzheimer's disease (AD). We wanted to find out if atrophy of the medial temporal lobe could be detected in life in patients in whom a diagnosis of AD was subsequently established histopathologically. The minimum width of the medial temporal lobe, measured by temporal-lobe-oriented computed tomography (CT) about one year before death, in 44 patients with a histopathological diagnosis of AD (cases) was nearly half (0.56 of the median) that in 75 controls of the same age with no clinical evidence of dementia (95% confidence interval 0.51-0.61). There was little overlap between the distributions of measurements in cases and controls. A cut-off (< 0.79 MoM) selected to yield a 5% false-positive rate gave an expected detection rate of 92%. A cut-off selected to yield a false-positive rate of 1% (< 0.70 MoM) yielded a 79% detection rate. 20 of the 44 patients with histopathologically diagnosed AD had been scanned more than once before death, and the test (cut-off < 0.79 MoM) was positive in all 20 more than a year before and in 9/10 more than 2 years before death. In 10 subjects with dementia but with histopathology excluding AD, the mean minimum width of the medial temporal lobe was significantly greater than that in the cases with AD, but was not significantly different from that in controls. Medial temporal lobe CT is a non-invasive, rapid, simple and effective test for AD which could have immediate application firstly in improving the accuracy of prevalence and incidence studies and, secondly, for the identification of groups of high-risk patients in the evaluation of novel treatments for AD. In the future, it could be applied as a screening test.