We assessed the accuracy of brain perfusion single-photon emission computed tomography (SPECT) in discriminating between patients with probable Alzheimer's disease (AD) at the very early stage and age-matched controls before and after partial volume correction (PVC). Three-dimensional MRI was used for PVC. We randomly divided the subjects into two groups. The first group, comprising 30 patients and 30 healthy volunteers, was used to identify the brain area with the most significant decrease in regional cerebral blood flow (rCBF) in patients compared with normal controls based on the voxel-based analysis of a group comparison. The second group, comprising 31 patients and 31 healthy volunteers, was used to study the improvement in diagnostic accuracy provided by PVC. A Z score map for a SPECT image of a subject was obtained by comparison with mean and standard deviation SPECT images of the healthy volunteers for each voxel after anatomical standardization and voxel normalization to global mean or cerebellar values using the following equation: Z score = ([control mean]-[individual value] )/(control SD). Analysis of receiver operating characteristics curves for a Z score discriminating AD and controls in the posterior cingulate gyrus, where a significant decrease in rCBF was identified in the first group, showed that the PVC significantly enhanced the accuracy of the SPECT diagnosis of very early AD from 73.9% to 83.7% with global mean normalization. The PVC mildly enhanced the accuracy from 73.1% to 76.3% with cerebellar normalization. This result suggests that early diagnosis of AD requires PVC in a SPECT study.