We evaluated the methodological issues underlying the assessment of normal confidence intervals, as used in clinically based region-of-interest (ROI) semi-quantification of 99Tcm-HMPAO brain SPET. At two different centres equipped with high-resolution, triple-headed gamma cameras, HMPAO SPET scans were performed on two groups of 24 and 15 healthy volunteers respectively. Together with an operator-defined analysis (ODA), a semi-automated analysis (SAA) was conducted on the normal datasets in one centre. Tests of intra- and inter-observer variability were performed. Repeat scans were performed within 72 h after the first to analyse short-term regional inter-study variations. The overall regional uptake showed significant differences in most regions between both normal datasets. Intra-observer and inter-observer reproducibility were on average within 4% for the ODA, while for the SAA it was less than 1%. Inter-study variations were excellent for both centres, ranging from -4% to +3% for most regions studied. The variability in clinical brain perfusion studies largely depends on the reproducibility of the data analysis technique. A semi-automated approach shows clear advantages over an entirely operator-defined approach. Intra-subject repeat studies show enough stability for use as reliable baseline measurements in the construction of a normal database or to allow activation studies with high sensitivity.