The paper presents an improved multiple signal classification algorithm based on coherent signals sources to estimate the direction of the arrival coherent signals. In the process of implement the algorithm, data eigen values are first decomposed from the received signal sources, and then using the relationship between the signal subspace corresponding to big eigen value and the signal steering vector to achieve effective estimation of the direction via searching the multidimensional wave peak. The algorithm is suitable for any arrays and the dimension of the covariance matrix will not be decreased. Compared with other algorithms, improved multiple signal classification algorithm has the advantages of owing the ability of processing up to M (M-1) signals and handling more than one signals from one direction. Results show that the proposed algorithm is suitable for applied in applications such as seismic detection and mobile communication.