Brain activation in studies using blood oxygenation level dependent (BOLD) FMRI is associated with an increase in T2* weighted signal between baseline and an active condition. This BOLD technique is often applied to study differences in brain activation between patients and healthy controls. However, the baseline T2* signal itself may also be different between groups, as shown in the hippocampus in Alzheimer's disease using the resting oxygen or ROXY approach (Small et al. : Ann Neurol 51:290-295). In the current study, we analyzed whole brain, voxel-wise T2* weighted signal of averaged baseline scans of a BOLD FMRI experiment in 41 healthy elderly controls and 46 patients with mild cognitive impairment or Alzheimer's disease. In each subject, T2* weighted images were normalized to the CSF signal of the same image. Additionally, gray matter probability maps of high-resolution structural scans were also compared between groups to assess atrophy. T2* signal was decreased in dementia in the hippocampus, insula/putamen, posterior and middle cingulate cortex, and parietal cortex. Most of these regions also showed decreased gray matter, except insula/putamen. Hippocampal and posterior cingulate gray matter differences were significantly larger than T2* differences. Therefore, decreased T2* signal in most regions are likely to be caused by gray matter atrophy, although decreased metabolism or perhaps iron deposition are also factors that may contribute. We conclude that in FMRI studies of dementia, not only the dynamic BOLD signal (activation and deactivation) but also the average baseline signal is diminished in certain regions. The method we applied may also be used in task-related BOLD FMRI and add to the understanding of the mechanism of task-related group differences.