For the segmentation of the non-brain tissues and brain tissues from the T2 weighted volumetric Magnetic Resonance Images (MRI), an image extraction method including two levels was proposed to extract brain tissues based on deformable surface models and mathematical morphology. The first level was finished by deformable surface models and region growing according to the brain anatomic, imaging knowledge, and the distribution of MR brain tissues in intensity histogram; the second level was completed by using mathematical morphology to erode the outcome of the first level of brain extraction to obtain more precise results. The experimental results demonstrate that the accuracy rate of the human brain extraction from T2 weighted MRI can reach more than 94%. The algorithm performance evaluation proofs that the proposed method is effective on the extraction of brain tissues from T2 weighted MRIs.