In this paper, the comparison between several numerical forecasting models for the case study of cold wave is described. The result shows, the 5-layer model^[1] for the NG model considering both the nonadiabatic and topographic effects is capable of predicting the crucial transferm of general circulation at the earlier stage of a cold wave.Its application to the 48-hr forecast at 500mb over Tibetan Plateau and its surroundings gives fewer absolute deviation values than the Japanese model^[2]. The MG model, which considers the topographic influence but results merely in a large scale condensation, is not so efficient as compared with NG, indicating that the radiation heating by long and short waves and the sensible heating are very important in weather forecast.The less importance of large scale condensation may be verified through a conparison between MG and the dry adiabatic or DG models. In the NN model, the nonadiabatic process is involved while the topographic effect is ignored.Its trial test for the 48hr forecast at the west of Tibetan Plateau leads to approximately the same result as NG model. However, this model when used for the forecast over the plateau and its neighbouring regions or at the front and the south of a trough gives a result not as good as NG.This shows that the forecasting quality differs very much depending on topographic conditions.