The China Meteorological Administration Numerical Weather Prediction Centre has established the GRAPESREPS (Global and Regional Assimilation and Prediction Enhanced SystemRegional Ensemble Prediction System) business system since 2014. The lateral boundary perturbation is obtained by the global ensemble prediction system. In order to understand the impact of lateral boundary perturbation on GRAPES regional ensemble prediction, this paper, based on the GRAPESREPS regional ensemble prediction model with 15 km horizontal resolution, constructed two lateral boundary perturbation schemes using the scaled lagged average forecasting (SLAF) method and the dynamic downscaling method. Two kinds of lateral boundary perturbation experiments were designed. The 6 d ensemble prediction test was carried out in July 2015. Using root mean square error (RMSE), spread, CRPS, Outlier, Brier score, TS, AROC and other probabilistic forecast test, the impacts of the two kinds of lateral boundary perturbation methods on the regional ensemble prediction quality were analyzed. The results showed that the energy of the lateral boundary perturbation obtained by dynamic downscaling (DOWN) is greater than that of the SLAF experiment at each vertical layer, resulting in the spread of DOWN experiment greater than the dispersion of SLAF experiment. For the isobaric elements and the ground elements, the scores of DOWN experiment are better than the scores of SLAF experiment, indicating that the lateral boundary perturbation of the DOWN experiment structure is more reasonable. In the aspect of precipitation probability prediction skill, the SLAF experiment has certain advantages in scoring, but the improvement of the score does not pass the significance test, so the two experiments are considered to be similar in the improvement of precipitation forecast.