Cerebral blood flow (CBF) can be quantified non-invasively using the brain perfusion index (BPI) determined from radionuclide angiographic data generated with 99mTc-hexamethylpropylene amine oxime (99mTc-HMPAO). When measuring the BPI, manual drawing of regions of interest (ROIs) (manual ROI method) for the extraction of the arterial input function (AIF) can lead to serious individual differences. The purpose of this study was to apply the fuzzy c-means (FCM) clustering method to determine AIF, and to investigate its usefulness in comparison with the manual ROI method. Radionuclide angiography was performed using a bolus injection of about 555 MBq of 99mTc-HMPAO, followed by sequential imaging (1 sec/frame x 120 s) using a solid-state gamma camera, and the BPI values were calculated using spectral analysis. To investigate the dependence of BPI on the ROI size, we drew five ROIs with different sizes over the aortic arch, and calculated the BPI using the manual ROI method [BPI(manual)] and the FCM clustering method [BPI(FCM)]. Furthermore, we asked 10 individuals to draw ROIs to investigate the inter-operator variability of the two methods. The mean and standard deviation (SD) of BPI(manual) increased with increasing ROI size, whereas the mean of BPI(FCM) was almost constant regardless of the ROI size; the SD of BPI(FCM) was smaller than that of BPI(manual). The inter-operator variability of the FCM clustering method was smaller than that of the manual ROI method. These results suggest that the FCM clustering method appears to be useful for the measurement of BPI, because it allows a reliable and objective determination of AIF.