Abstract: Performance of direction-of-arrival (DOA) estimation decreases obviously in the actual ocean ambient noise. To tackle this problem, the real-part elimination of covariance matrix method is proposed. Firstly, the noise field is decomposed into symmetrical noise filed and asymmetrical noise field. Secondly, the real part of covariance matrix is eliminated so as to suppress the symmetrical noise, and then the DOAs are estimated using imaginary part and delay-and-sum beamforming method. Meanwhile, it produces false targets. Finally, a new covariance matrix, which is composed of the imaginary part and the reconstruction of the real part, is applied to estimate DOAs so as to eliminate the false targets. Both the theoretical analysis and experimental results show that this method improves performance of direction-of-arrival estimation by suppressing noise.