One hundred and thirty-eight patients satisfying NINCDS/ADRDA criteria for Alzheimer's disease (AD) and 36 normal controls underwent cranial CT. A comparison of methods of scan analysis showed good agreement between computer-assisted methods and visual ratings and planimetry. The CT scans of controls differed significantly from patients and a discriminant analysis, based on all CT measures, predicted group membership (control or patient) in 81% of cases. Within the AD group, cortical atrophy correlated with age and duration of illness. Global tests of cognitive function correlated significantly with both cortical atrophy and ventricular size. Subjects who died in a three-year follow-up had more atrophy and larger third ventricles than survivors, but this effect was due entirely to increased age.