For strongly correlated or coherent signal sources, the high-resolution methods based on subspace decomposition can not accurately estimate signal direction of arrival. While most decorrelation algorithms, such as spatial smoothing, will reduce the efficient aperture, and is only suitable to the array structure with shift-invariance. A new DOA estimation algorithm based on projection matrix search is proposed for coherent sources. Firstly, the projection matrix of the noise subspace is constructed, which is based on the array manifold, and then the array received signal is projected into the noise subspace. Secondly, the spatial spectrum is calculated by searching all possible projection matrices, thus the DOA estimation of coherent sources is obtained. The proposed algorithm can effectively estimate the DOA of coherent sources. Compared with the conventional DOA estimation method for coherent sources, the proposed algorithm will not reduce array aperture, and can be applied to any arrays, and possesses better estimation accuracy and super-resolution capability, but the computational complexity of the proposed algorithm is higher. Simulation results show the effectiveness of the proposed algorithm and its superior performance over conventional algorithms.