As the source number estimation algorithm based on information theoretic criteria cannot adapt to the correlated source,this paper proposes a new one based on an improved spatial smoothing .The algorithm recombines the subarrays so that each subarray has the same number of elements with the original array. The study takes arithmetic mean of every subarray covariance matrix to form a new array output covariance matrix,then estimation source number with information theoretic criteria based on output covariance matrix. Simulation and analysis show that the algorithm can correctly estimate the source number in the presence of the corrected source .It can also improve the utilization rate of the array elements.The performance is better than the algorithm of information theoretic criteria based on spatial smoothing in low SNR and number of snapshots.