The shape, size and location of regions of interest (ROIs) show considerable variability between single-photon emission tomography (SPET) studies in aging and Alzheimer's disease, but the possible influence on study results remains unknown. We compared three different ROIs in a SPET study with 60 controls and in 48 patients with probable Alzheimer's disease diagnosed according to the NINCDS-ADRDA criteria. Regional cerebral blood flow (rCBF) was assessed with SPET using technetium-99m d,l-hexamethylpropylene amine oxime (99mTc-HMPAO), normalized to the mean activity in a cerebellar reference slice. The three different ROIs were: a multi-slice and a single-slice ROI with reference to the normal brain anatomy (using an anatomical atlas), and a rectangular (2 x 4 pixels) ROI in the frontal, temporal, temporoparietal and occipital cortices. No differences were observed for the means of rCBF values between the single-slice and multi-slice ROI's with reference to the normal anatomy, but some variability was present for individual comparisons. In contrast, significantly higher mean rCBF values were obtained with the single-slice rectangular ROIs in all four regions for both patients and controls and considerable variability was shown for individual subjects. After analysis with multivariate logistic regression and receiver operator characteristic curves, the ability of SPET to discriminate between controls and Alzheimer patients was similar in the three methods for mild and moderate Alzheimer patients (Global Deterioration Scale = GDS of 3 and 4). However, with increasing dementia severity (GDS > 4) the rectangular ROIs showed lower ability to discriminate between groups compared to the single-slice and multi-slice anatomically defined ROIs.(ABSTRACT TRUNCATED AT 250 WORDS)