We used magnetic resonance imaging (MRI) measurements to determine whether persons in the prodromal phase of Alzheimer's disease (AD) could be accurately identified before they developed clinically diagnosed dementia. Normal subjects (n = 24) and those with mild memory difficulty (n = 79) received an MRI scan at baseline and were then followed annually for 3 years to determine which individuals subsequently met clinical criteria for AD. Patients with mild AD at baseline were also evaluated (n = 16). Nineteen of the 79 subjects with mild memory difficulty "converted" to a diagnosis of probable AD after 3 years of follow-up. Baseline MRI measures of the entorhinal cortex, the banks of the superior temporal sulcus, and the anterior cingulate were most useful in discriminating the status of the subjects on follow-up examination. The accuracy of discrimination was related to the clinical similarity between groups. One hundred percent (100%) of normal subjects and patients with mild AD could be discriminated from one another based on these MRI measures. When the normals were compared with the individuals with memory impairments who ultimately developed AD (the converters), the accuracy of discrimination was 93%, based on the MRI measures at baseline (sensitivity = 0.95; specificity = 0.90). The discrimination of the normal subjects and the individuals with mild memory problems who did not progress to the point where they met clinical criteria for probable AD over the 3 years of follow-up (the "questionables") was 85% and the discrimination of the questionables and converters was 75%. The apolipoprotein E genotype did not improve the accuracy of discrimination. The specific regions selected for each of these discriminations provides information concerning the hierarchical fashion in which the pathology of AD may affect the brain during its prodromal phase.