Abstract: Persistent heavy rainfall events (PHREs) are difficult to forecast. Medium-range objective weather forecast technology for PHREs based on the global ensemble prediction system (EPS) of the China Meteorological Administration (CMA) is constructed based on the Anderson-Darling test principle. Using T639 forecasts of precipitation and observed rainfall during May-September of 2010-2015, characteristics of the cumulative distribution functions (CDFs) of the observed and T639 EPS forecasts of precipitation are analyzed. Accordingly, three schemes of modeling climatological CDFs are established using the method of extending time series and spatial range. Through a batch of forecasting tests, the optimal modeling scheme is selected for application in the forecast model to examine the longest valid forecast time for PHREs. The result shows that the precipitation forecasted by T639 EPS is gradually concentrated in light and moderate rain with increase in the forecast lead time, and the probabilities of no precipitation and rainstorm both are lower than observations. The probability of precipitation is stable after 168 h. The scheme can make up for the deviation of EPS data by extending the time series and the spatial range. Further subdivision of modeling of climatological CDFs according to regional climatic characteristics highlights the precipitation characteristics over different areas, which is significantly better than the one that simply incorporates all regional data into one single dataset. The model based on T639 EPS can issue early warnings of PHREs 8 to 9 d in advance. With increases in the forecast lead time, the EPFI becomes less skillful in forecasting rain belt and intensity.