This paper presents an approach of bas-relief stylization based on normal field. To this end, our method reconstructs the resulting bas-relief with the altered local details by solving a global optimization. First, Gaussian filters are used to extract the high-frequency information and low-frequency information from an initial height image. Next, the separate height images are respectively transformed into two normal images by a bilateral surface fitting method. The separate normals are edited respectively using different stereographic mapping methods. Finally, a bas-relief is generated by solving a sparse linear system in the least-squares sense. By extracting different frequencies from a height field and editing them on normal do-main, our approach naturally preserves and enhances the fine details of model and produces bas-reliefs with different styles. The experimental results show that our approach is easy to implement, and effectively demonstrate its ability to restyle digital bas-reliefs.