The CMA T213 global ensemble prediction system using BGM initial perturbation scheme has not considered model perturbation thus lags behind the ensemble forecasting system of international advanced technical centers. This paper referring to the ECMWF model perturbation method, designs a T213 global ensemble prediction system stochastic physics perturbation method, and conducts ensemble prediction tests in 20-31 July 2008. The results show that, the T213 global ensemble prediction system is very sensitive to stochastic physics perturbation. This is because after physics perturbed, predictor variables change significantly, and the changes expand rapidly with the integration time of growth. In the horizontal direction, the middle and high latitudes are more sensitive than the equatorial regions. In the vertical direction, the variables characterizing the largescale movements, such as geopotential height, temperature, and wind speed are very sensitive from low to upper levels, and the most sensitive is at 300 hPa, in the middle and high latitudes of north and south hemispheres; while vertical velocity, divergence and other physical variables in the equatorial region are also very sensitive. After the multiple initial condition ensemble added to the stochastic physics perturbations, the spread and RMSE of ensemble mean are improved slightly in the late term of integration, while the improvement of the precipitation forecast is significant, which indicates the prospect of operation to the stochastic physics perturbation is good. The next step will be more test assessments, and the operations of the stochastic physics perturbation are as early as possible to shorten the distance between China and the international advanced technology in the ensemble.