Abstract:The performance of medium range forecasts is verified and compared for the T639, ECMWF and Japan models from June to August 2017. The results show that all of the three models could well predict the variation and adjustment of the atmospheric circulation over middle and high latitudes in Eurasia, of which ECMWF model performs the best. ECMWF and T639 models have good performance in predicting activity of western Pacific subtropical high (WPSH), and the ECMWF model shows better skill. The prediction bias of the west ridge of WPSH reduces from early summer to late summer. The T639 and ECMWF models underestimates the temperature over most of China at 850 hPa, and the JP model overestimates the temperature over most of China. In southern China, three models have smaller bias of temperature prediction than that in northern China. ECMWF model has better performance than T639 and Japan models. As far as Typhoon Hato (1713) is concerned, ECMWF model performs better in predicting the track, intensity and landing position, although the predicted strength is weaker. The T639 model shows a poor skill in the prediction of Typhoon Hato.