In order to reduce the influence of mutual coupling on DOA estimation, this paper proposes a weighted sparse representation method for DOA estimation with unknown mutual coupling. The proposed method divides the space into grids. Each grid is 1° partitioned. We utilize the Capon spectrum to design a weighted weights with unknown mutual coupling and propose to use the SVD of the data matrix to mold the problem with reduced dimensions. At last, we estimate the angle of the signal based on the L1norm. Simulation results demonstrate that our proposed method can obtain better DOA estimation result and lower RMSE. Compared with traditional methods, the new algorithm can effectively suppress the unknown mutual coupling.