Based on the European Centre for Medium-Range Weather Forecasts(ECMWF),National Centers for Environmental Prediction(NCEP) and the China Meteorological Administration(CMA) ensemble forecasts of 2 m surface air temperature taken from TIGGE data archive,the inconsistency of the surface air temperature has been investigated.The results show that the time mean inconsistency index of each model for surface air temperature is quite different.The time mean inconsistency index of ECMWF forecast system is the smallest,followed by NCEP,and CMA is the largest.In addition,the inconsistency indices of the control forecast in the NCEP forecast system,the control forecast and the ensemble mean forecast in the ECMWF forecast system increase with the forecast lead time.In the ensemble forecast system,the ensemble mean forecast is superior to the control forecast,especially for longer term forecast.But in the CMA forecast system,the inconsistency indices of the control forecast and the ensemble mean forecast are substantially stable at a higher level.The further study indicated that the time mean inconsistency index of the surface air temperature forecast in the ECMWF ensemble forecasting system varies with areas and seasons.The forecast inconsistency index during winter(summer) is relatively larger(smaller) among four seasons in a year.Furthermore,the occurrence frequency of the forecast flip doesn't vary with forecast lead time significantly,while that of the forecast flip-flop and forecast flip-flop-flip increases with the forecast lead time considerably.