Summary: A novel approach to the simplification of dynamic systems is introduced. It is suggested that a high-order linear system can be approximated by a low-order model with a time delay based on the proposed mixed method. The denominator of the reduced model is determined by the Routh stability method, which gives rise to the desired stability of the model produced. The numerator parameters are then estimated by using identification techniques and the time delay is chosen optimally according to an output- errors criterion. It is shown that the reduced-order time-delay model produces a good enough approximation compared with models obtained from other methods. The reduction procedure is fully computer oriented. The performance of the new approach is verified by the application of numerical examples.