Accurate segmentation of substantia nigra (SN) and red nucleus (RN) in quantitative susceptibility mapping (QSM) images has great clinical value in quantifying iron deposition and measuring disease severity. We propose a new segmentation algorithm which uses the discontinuity of seed points in different tissues as prior knowledge. Seed points in SN or RN can be obtained from standard atlas or specified manually. This prior was then incorporated into level set method to segment SN and RN. Experiments on in-vivo MR images showed that the proposed method achieved more accurate segmentation results than the atlas-based method and classic level-set method.