The medium range forecast products of the numberical models T639, ECMWF and Japan from December 2012 to February 2013 are compared and verified. The results show that the three models have good performance in predicting the variation and adjustment of atmospheric circulation over Asian middle and high latitude areas, but all of them have systemic errors increasing with forecasting time. Comparatively speaking, EC model has better performance than T639 and JP models. For the temperature field at 850 hPa, EC model has better performance than JP in the north of China but in the South, JP plays better than EC. Comparatively, the forecast errors of T639 model are the biggest no matter in the south or the north of China. For the process of weak cold airs in the Northeast of China, JP model has the best performance, T639 model produces obvious errors in predicting the intensity and position of cold high, and EC also creates some small errors in intensity and position forecasts.