In order to surmount the disadvantage that much time is consumed for feature extraction with traditional K L transform, an improved K L transform method in pattern recognition is presented. This improved method gets covariance matrix with recursion algorithm and computes the eigenvalues and eigenvectors of the covariance matrix with matrix perturbation theory. The simulation results show that compared with the traditional method, the consuming time with this method is sharply decreased, and the computing precision is kept.