Acoustic environment has a low signal-to-noise ratio (SNR); hence, array signal processing is widely used for noise reduction and signal enhancement. The actual ambient noise includes uncorrelated noise and correlated noise. The received noises of the two arbitrary array elements are correlated. Consequently, the performance of array signal processing method decreases obviously. Aiming at this problem, the real-part elimination of covariance matrix method is proposed. Firstly, from a physical point of view, the noise signals can be generated by using a number of uncorrelated noise sources: the more the noise sources, the less the error between the noise from the model and the actual noise will be. Theoretically, the noise field is decomposed into the symmetrical noise field and the asymmetrical noise field. A number of noise sources generate the symmetrical noise fields; the directions of these noise sources are symmetric, and the powers of two arbitrary symmetric sources are the same. Secondly, the symmetry of the ambient noise is analyzed, as a result, the symmetrical noise can only affect the real part of the covariance matrix. Thirdly, the real part of covariance matrix is eliminated in order to reduce the noise, and then the delay-and-sum beamforming is achieved by using only the imaginary part. The advantages are that the output signal-to-noise ratio is increased and the noise output power is reduced obviously; the disadvantage is that it produces a false target. The azimuth of the actual target differs from that of the false target by 180, and the false target cannot be distinguished. Finally, to eliminate the false target, the real part of the signal covariance matrix is reconstructed by establishing a constrained optimization problem, which is solved by using the particle swarm algorithm. Then, the reconstructed covariance matrix composed of the imaginary part and the reconstruction of real part is applied to delay-and-sum beamforming, as a result, the false target is eliminated. The simulation results show that the real-part elimination of covariance matrix method reduces the symmetrical ambient noise, the noise output power is reduced, the output signal-to-noise ratio is increased, and this method improves the performance of array signal processing. The experimental results show that the output SNRs of two targets with using the imaginary part of covariance matrix are increased by 3.57 dB and 3.149 dB, respectively, and the output SNRs of two targets with using the reconstructed covariance matrix are increased by 7.027 dB and 6.985 dB, respectively. The real-part elimination of covariance matrix method is easy to implemente, and has a definite value for engineering application.