We studied regional cerebral blood flow (rCBF) by xenon-133 dynamic single photon emission computed tomography (SPECT) in 17 normal volunteers who were performing the Wisconsin Card Sorting Test (WCST), a task that is particularly sensitive to disturbance of the prefrontal cortex, and a simple matching-to-sample task (BAR) as a sensorimotor control. Three methods for statistical analysis of regional "subtraction" data were used: absolute rCBF values, percent distribution values, and means adjusted for global CBF changes (covariance analysis). The absolute values had high variance, due to the combination of interindividual differences in global flow and intra-individual variation, and showed no statistically significant regional changes. This variation was greatly reduced by percent values and covariance analysis, which had quite similar outcomes. With both methods, significant increases of rCBF during the WCST as compared with the BAR were seen in the right anterior dorsolateral prefrontal and left occipital cortices, and reduction of rCBF in the left pararolandic region. Moreover, significant correlations with performance were found in the medial regions of the frontal lobes, with opposite trends for the right and left hemisphere. The posterior dorsolateral prefrontal region showed a negative correlation with sensory-motor frequency, an index related to the task's difficulty. These results are consistent with previous findings using other rCBF techniques and confirm the statistical advantage of normalization and covariance methods, which yield practically identical results, at least in this analysis based on regions of interest.