The course of multiple sclerosis (MS) can be monitored by measuring changes in brain volume, but consensus is still lacking on the best strategy to be adopted. We compared the reproducibility and sensitivity of volume measurements from different brain portions for detecting changes on magnetic resonance imaging (MRI) in patients with MS. T1-weighted MRI of the brain was performed in 50 patients with relapsing-remitting MS at study entry and after an average follow-up of 18.4 months. Using a semiautomated technique for brain parenchyma segmentation, the volumes of the following brain portions were measured: (a) the whole brain (whole-brain volume, WBV), (b) the seven slices rostral to the velum interpositum (seven-slice volume, SSV), (c) the central slice of the image set (central-slice volume, CSV) and (d) the infratentorial regions (infratentorial-brain volume, IBV). All these measurements were carried out by a single observer and were repeated twice on ten randomly selected scans to test the intra-observer reproducibility using the four strategies. At follow-up there was a significant decrease in all the measures of brain volume (P ranged from 0.002 to < 0.001). The univariate correlations between changes in WBV, SSV, CSV and IBV were all statistically significant, with the exception of that between changes in CSV and IBV; r values ranged from 0.34 (for the WBV/IBV correlation) to 0.80 (for the WBV/SSV correlation). The mean intra-observer coefficient of variations were 1.9% for WBV, 1.5% for SSV, 2.9% for CSV and 2.2% for IBV measurements. The measurement of volume on a portion of brain selectively including the regions in which MS pathology is more diffuse is as reliable and sensitive to disease-related changes as that on the whole brain, with significant time saving for processing.