It is hard to segment the magnetic resonance image which has Gaussian noise. In this paper, we presented a novel approach for segmentation of brain MRI data. It could improve the accuracy and robustness of segmentation. First of all, a new energy function based on characteristics of brain data and Gaussian noise distribution was derived. Then, in order to obtain the stable segmentation results, active contour model and level set method was introduced into the energy function. Finally, segmentation results were achieved by minimizing the novel energy function. Experiment results show that the presented method is valid. Compared with the traditional algorithms, the new method has higher accuracy to the T1_weighted brain MRI image. The accuracy rate of gray matter, white matter and CSF of new algorithm is high 7.1%, 7.5% and 22%. For visual quality, the proposed method can distinguish the similar regions effectively and reduce “granule”.