In this methodological study, a procedure for measuring regional CBF (rCBF) with positron emission tomography and 15O-labelled tracers is optimized. Four healthy volunteers were subjected to eight studies with use of [15O]butanol as a tracer: four times while reading aloud and four times while reading silently from a phonologically balanced list of single words. The gain from these repeated intra-individual studies of the same activation state (fractionation) was demonstrated in terms of noise-equivalent counts in a phantom study. A computerized brain atlas was used to reformat the images to a common anatomical representation, thereby minimizing the effects of inter- and intra-individual anatomical and positional variations. This allowed the formation of inter- and intra-individual average subtraction images with error estimates. Differences between the two activation states were detected with use of an exploratory significance map based on a paired Student's t test. The results compared well with Friston's method of determining levels of statistical significance. No difference was obtained when comparing results from rCBF images and images generated from measurement of uptake of the tracer. The paradigm chosen for activation was shown to yield a constant activation level during the repeated measurements (i.e., no habituation).