Fourteen patients who had clinically diagnosed Alzheimer's disease with mild to severe dementia (mean age 69.1 years) were evaluated by calculation of local cerebral metabolic rate for glucose (LCMR-gl) based on uptake of 18F-2-fluoro-2-deoxyglucose (FDG) detected with positron emission tomography (PET). PET scanning showed that the patients had significantly lower LCMR-gl values than 11 age-matched neurologically normal volunteers (mean age 66.3 years). The differences were most marked in the temporal cortex, followed by the frontal, parietal and occipital cortex. In each case the LCMR-gl value was below the lowest control value in at least one cortical area and usually in several; the reduction in LCMR-gl and the number of regions involved in the patients increased with the severity of the dementia. Deficits noted in neuropsychologic testing generally correlated with those predicted from loss of regional cortical metabolism. The patients with Alzheimer's disease were also examined with magnetic resonance imaging, computed tomography or both; the degree of atrophy found showed only a poor correlation with the neuropsychologic deficit. Significant atrophy was also noted in some of the controls. A detailed analysis of LCMR-gl values in selected cerebral regions of various sizes refuted the hypothesis that the reduction in cortical glucose metabolism in Alzheimer's disease is due to the filling by metabolically inert cerebrospinal fluid of space created by tissue atrophy.