A rainstorm process in South China during the pre-rainy season of 2006 has been selected in this article. AREMv2.3 mesoscale numerical model is used to simulate the process through adding perturbations to the initial fields of physical variables (temperature, humidity, and wind). The authors analyze the impacts of perturbations of different physical variable fields on the precipitation forecast, the error growth of physical variables, and the perturbation energy growth. And the authors also discuss the relationship between the error growth and moist convection, and the impact of perturbations of different amplitudes on the error growth, and the predictability of the mesoscale precipitation in South China. The numerical simulation results show that: the impacts on precipitation are different when adding the perturbations of the actual amplitude normal distribution at initial time to different physical variable fields. For 24-hour precipitation forecast, the temperature perturbation has the greatest impact on precipitation. Error growth and moist convective instability are closely related. The small-scale and small-amplitude initial errors increase rapidly, furthermore in the form of nonlinear growth. This means that there is less predictability in short-term smaller-scale precipitation forecast. Compared with the larger-amplitude perturbations, smaller-amplitude perturbations can cause smaller error. But smaller-amplitude perturbations develop rapidly and will soon make a great impact on the precipitation forecast. Therefore, it can only improve the prediction quality limitedly, and because the perturbations are in quick nonlinear growth, there will not be much improvement that can be done to bring forward the prediction time.