Abstract: Based on the WRF model, in this study, the authors investigated the source of forecast errors for an extreme precipitation event in the southwest of Guangdong Province that occurred from 16 May to 17 May 2015. First, forecasts using the WRF model with the initials NCEP_FNL (hereinafter WRF_FNL) were compared with the deterministic forecasts generated by ECMWF (European Centre for Medium-Range Weather Forecasts), and the results showed that ECMWF had good forecast skills. Thus, the model and initials used by ECMWF are deemed to be accurate. Then, the authors generated a new forecast (hereinafter WRF_EC) using the WRF model with the ECMWF initials, keeping the same physical schemes as those in the previous WRF simulations. The results showed that WRF_EC has better forecast skills than WRF_FNL. This indicates that the improvement of initial states has a significant impact on forecasts, and thus the initial error is the main source of the forecast errors. The authors also analyzed the sensitive area and sensitive variables. The results showed that the initial errors mainly came from the North Gulf of the South China Sea and southwest of Guangxi Province. Both the initial temperature and the initial humidity played an important role in the heavy precipitation forecast. Thus, improving the accuracy of initial temperature and humidity could greatly improve the skills in forecasting this heavy precipitation phenomenon.