This paper presents a method of blind source separation (BSS) of autoregressive (AR) signals for complete and over-complete cases. It is based on a new separation matrix estimated from the null space (NS) of the mixture. Analysis of a mixture equation is carried out to find out the analytical representation of the separating matrix used for estimating the input source signals. The Eigenvectors of the mixture matrix is factorized using upper and lower triangular matrix factorization, then use them to formulate the separation matrix. An algorithm of the new method is provided in this paper. Simulation results show that the method is successfully separating speech and Gaussian signals from their mixture with MSE less than 0.14.