Based on the ensemble forecasting data of China Meteorological Administration(CMA),European Centre for Medium Range Weather Forecasts(ECMWF),Japan Meteorological Agency(JMA),U.S.National Centers for Environmental Prediction(NCEP),and United Kingdom Met Office(UKMO) in the TIGGE datasets,the multimodel ensemble forecasting techniques have been investigated.Results show that the multimodel ensemble forecasts are superior to that of the individual model.However,the improvement of multimodel ensemble forecast skill is different for different meteorological elements.The multimodel superensemble with running training period(R-SUP) has the best performance for surface air temperature in the Northern Hemisphere.For precipitation forecast in the Northern Hemisphere,the bias-removed ensemble mean(BREM) is more skillful and stable than every single model.In addition,the multimodel ensemble forecasting experiments of extreme weather event with freezing rain and snow over southern China during early 2008 have been conducted by using BREM with running training period(R-BREM).Taking root-mean-square errors(RMSEs) of ECMWF 96 h forecasts as the criterion,the forecast lead time of surface air temperature over the southeastern coast of China may be prolonged from 96 h to 192 h by using the R-BREM technique.And the threat scores(TS) of the light and moderate rain forecasts have been significantly improved except for some particular lead time.