Regarding the forecasting errors of operational models for the high-impact extremely heavy rainfall event in Beijing on July 21, 2012, this work examines the impact of assimilating surface and rawinsonde observations using EnKF data assimilation system on the simulation of rainfall distribution and the surface features in the initiation period of the rainfall in Beijing, and reveals the possible reasons for the forecasting errors. Results show that data assimilation significantly improves the simulation of rainfall distribution, confirming that the cyclonic vortex is the key influencing system of the heavy rainfall event, which was proposed by previous researchers based on observations and sensitivity analyses. This work also reveals that the surface low and its associated inverted trough are the direct producers of the rainfall in Beijing. These results indicate that the reason of the failure of the operational models in this extremely heavy rainfall is the large forecasting errors in the strength and location of the cyclonic vortex and the associated inverted trough eastward of the surface low.