We used MRI volume sampling with coregistered and atrophy corrected FDG-PET scans to test three hypotheses: 1) hippocampal formation measures are superior to temporal neocortical measures in the discrimination of normal (NL) and mild cognitive impairment (MCI); 2) neocortical measures are most useful in the separation of Alzheimer disease (AD) from NL or MCI; 3) measures of PET glucose metabolism (MRglu) have greater diagnostic sensitivity than MRI volume. Three groups of age, education, and gender matched NL, MCI, and AD subjects were studied. The results supported the hypotheses: 1) entorhinal cortex MRglu and hippocampal volume were most accurate in classifying NL and MCI; 2) both imaging modalities identified the temporal neocortex as best separating MCI and AD, whereas widespread changes accurately classified NL and AD; 3) In most between group comparisons regional MRglu measures were diagnostically superior to volume measures. These cross-sectional data show that in MCI hippocampal formation changes exist without significant neocortical changes. Neocortical changes best characterize AD. In both MCI and AD, metabolism reductions exceed volume losses.