This paper is on the generalized singular perturbation approximation (GSPA) preserving the discrete positive real property. We transform the discrete positive real(PR) system into a stochastically banlanced system and get the reduced order discrete system from the GSPA of the full order stochastically balanced system. eSPECIALLY, WHEN THE FREE PARAMETER OF THE gspa IS .+-.1, we show that the reduced order discrete system retains stability, minimality, and positive real and stochstically balancing properties. And we derived the .inf.-norm error bound with the reduced order discrete strictly positive real(SPR) system by the proposed method. Finally, we give an example to ascertain the properties of the proposed reduced order discrete system and to compare with the conventional methods.