Abstract: Upon using the dynamic statistics of the at mospheric circulation the monthly mean forecasting fields of January and July of 1996 obtained from the second generation medium-range numerical weather prediction(NWP) model(T63) have been investigated. The results indicate that the new generation model has been improved remarkably in comparison with the old one. The systematic prediction errors of model have been reduced significantly. However there still exist some differences in the distributions in wind, temperature and moisture fields between forecasts and observation. After having diag nosed the transfers and budgets of angular momentum, heat and water vapour of the T63 model, and compared these with observ ations, it was suggested that, to improve the model performance, the treatments of orography, land surface processes, cloud radiation and convection paramet erization in the model need to be improved further.