Abstract: This paper presents a new method of searching optimal reduced-order model for a discrete time system.An error polynomial is defined,the coefficients of which indicate the difference at any instant between the sysem and the model.It is shown how Markov parameters and H-parameters of the model can be matched with those of the system by setting error polynomial coefficients to zero.The H-parameters are proportional to the system time moments.The searching of optimal matching of Markov and H-parameters are based on one of seven various integral of the error performance index that minimizes it.This program is written in PASCAL language and runs on a PDP-11/23 minicomputer.By taking an actual computer system as an example we achieved simulation and comparison.It is shown that this method is quite correct,feasible,efficient,and easily understandable.