Abstract: In this paper, a novel high-resolution direction-of-arrival estimation method was presented based on sparse spectral fitting to overcome the drawback that the performance of iterative re-weighted least squares algorithm could be impacted with the overcomplete basis matrix condition number. A singular value decomposition (SVD) was employed to handle the overcomplete basis, adopting the truncated SVD (TSVD) method to remove those singular vectors that corresponded with smaller singular value and obtain a well-conditioned matrix, and using this matrix to replace the overcomplete basis matrix. Then a regularized FOCUSS algorithm with l_p norm constraint was applied for sparse signal reconstruction to resolve ill-posed problem when the overcomplete basis matrix condition number got too large, and coarse-refined space grid separation was used to decrease the computational complexity. Simulation results show that compared with MFOCUSS algorithm, the proposed method can not only reduce computational complexity, but also hold much higher resolution and robustness to noise.