To improve fine temperature forcasting, based on the result amending of MM5 by grey dynamic model, and using the temperature interpolation methods with altitude and distance weight, the high and low temperatures in 72 hours at 717 automatic weather stations with 4 meteorological elements in Hubei Province, are forecasted on the basis of the MOS technique and the mesoscale model and tested by observations. The results show that the test score in this method to the stations without historical data is better than that by the mesoscale model and is approximate to that by the MOS technique,〖JP2〗 thus this method can be used in actual operations. The test also indicates that the forecasting difficulty in northwestern Hubei is the most compared with other areas and is the least in the Jianghan Plain. In this method, the forecasting errors of the lowest temperatue in autumn and spring are more than those in other seasons, and the prediction errors are the smallest in summer. Meanwhile, the trends of low temperature are similar in five areas of Hubei. However, the change trends of maximum temperature forecasting are inconsistent in these areas.