In order to test the proposed method, we selected two public databases (IXI and RBVM) of structural MRI images of healthy adults and two commonly used brain extraction methods for evaluation, BET and FreeSurfer. The algorithm uses volume stability as a regulatory parameter, i.e., assisting to stop the brain frontier correction when the volume change between iterations reaches a stable level. This is achieved by an iterative evaluation criteria based on the discontinuity of the pixel signal level and its relationship among the local neighborhood, estimated by use of first and second momentum for a moving window centered at each voxel of the brain surface. The proposed method is able to refine a prior brain volume segmentation, eliminating local patterns that are inconsistent with the brain borders. This study presents an automatic brain volume refinement (BVeR) method. Although the skull stripping procedure is time-consuming for manual assessments, most automatic approaches often present minor segmentation errors, commonly requiring a manual refinement. One of the possible data preparation steps is the skull stripping, which removes non-brain tissues from the original image. In neuroimage studies, it is usual to set a sequence of image preprocessing steps to obtain a controlled data analysis.