The large number of false positives that result when automatic algorithms are considered for segmenting small multiple sclerosis lesions in magnetic resonance imaging hampers the posterior evaluation of lesion load. To address this problem we propose a fuzzy system which can improve the differentiation between true and false positive detections in proton density- and T2-weighted images. On the basis of an earlier work, which was focused on the detection of hyperintense regions in MR brain images, the system here presented introduces fuzzy restrictions derived from the regional analysis of the main features in such regions. Results show a reduction to a 3.6% in the number of false detections while preserving most of the true detections obtained using previous algorithm.