The spatial resolution of current positron emission tomography (PET) scanners does not allow a distinction between cerebrospinal fluid (CSF) containing spaces and contiguous brain tissue. Data analysis strategies which therefore purport to quantify cerebral metabolism per unit mass brain tissue are in fact measuring a value which may be artifactually reduced due to contamination by CSF. We studied cerebral glucose metabolism (CMRglc) in 17 healthy elderly individuals and 24 patients with Alzheimer's dementia using [18F]fluorodeoxyglucose and PET. All subjects underwent x-ray computed tomography (XCT) scanning at the time of their PET study. The XCT scans were analyzed volumetrically, in order to determine relative areas for ventricles, sulci, and brain tissue. Global CMRglc was calculated before and after correction for contamination by CSF (cerebral atrophy). A greater increase in global CMRglc after atrophy correction was seen in demented individuals compared with elderly controls (16.9% versus 9.0%, p less than 0.0005). Additional preliminary data suggest that volumetric analysis of proton-NMR images may prove superior to analysis of XCT data in quantifying the degree of atrophy. Appropriate corrections for atrophy should be employed if current PET scanners are to accurately measure actual brain tissue metabolism in various pathologic states.