Abstract: A new spatial spectrum estimation algorithm is presented.The basic idea of the algorithm is that only the smallest eigenvalue of R is a noise eigenyalue and so is the eigenvector.So the dimension of noise space is always one.The accuracy and the resolution.of estimation of DOA by the use of the new algorithm are higher than those from MUSIC in the actual condition that some energy of signals eigenvalues leak to noise eigenvalues.The new concept can also be used to modify other spatial spectrum estimation algorithms such as spatial smooth algorithm used for dealing with coherent sources.