Abstract: The performance of Direction-Of-Arrival (DOA) estimation of classical algorithms may degrade substantially in the presence of mutual coupling. In this paper, a blind source separation based algorithm is proposed for DOA estimation using uniform linear array. Firstly, the generalized array manifold matrix is estimated using blind source separation techniques; Then, based on the property that the mutual coupling matrix of uniform linear array can be modeled by a symmetric Toeplitz matrix, DOA estimation problem is transformed into multiple separable nonlinear least squares problems, and DOAs can be obtained by multiple one-dimensional peak searching. The proposed algorithm does not need multidimensional peak searching or multidimensional iteration, and imposes weaker constraints on the number of non-zero coefficients in the mutual coupling matrix. Besides, the proposed algorithm is applicable even if the number of non-zero coefficients in the mutual coupling matrix is unknown. Simulation results demonstrate the effectiveness of the proposed algorithm.