A self-organized Local Linear Embedding(SO-LLE) algorithm is proposed.This new algorithm can not only determine the value of the nearest neighbor automatically and decrease the operation quantity,but also can gain a perfect approximation of face manifold.Theoretical analysis is given in details and experiments are carried on some dataset to verify the effeteness of this method.The experimental results on Yale and PIE face databases show that the proposed algorithm can increase the average recognition rate from 5% to 40%,which improves face recognition performance effectively.