Abstract: Based on observational data from surface meteorological stations and radiosonde data,the forecast products,such as precipitation,geopotential height field at 500 hPa,subtropical high index,from different numerical models for the severe torrential rain event on August 16,2013 were validated and comparatively analyzed.The results show that general precipitation forecast accuracy of the the T639 model is better than that of the the EC model,and the mean forecast accuracy of rainstorm from the EC model is slightly better than that from the T639 model.The spatial distribution pattern of positive and negative precipitation anomaly of the T639 is similar to that of the EC.The precipitation predicted by different numerical models is all significantly smaller than observations during the main precipitation period (11:00-23:00 on August 16) at Qingyuan station,and the 3-h maximum precipitation predicted by the WRF model is much larger than the real precipitation.The precipitation grades predicted by the T639 and the EC are both smaller than the real grades.The 72-h precipitation area and the center position of strong precipitation predicted by the EC and an ensemble model are relatively stable.The 15-d moving averaged validating results have some referential meaning to the precipitation forecast.The consistency of 72-h characteristics line predicted by the EC is better than that predicted by the T639,and the dispersion of precipitation predicted by the EC over the most regions of Liaoning province as well as upper trough in the upstream region is smaller than that predicted by the T639.