Abstract: In order to use MM4 model successfully in the whole region of China, it is necessary to improve the cumulus convective heating parameters in MM4 model based on the statistical analysis of real climatic data in China. The routine radiosonde data in June~Aug, 1990~1993 are collected to calculate the vertical profile of convective heating directly at each station for each time level using Kuo’s scheme without unnecessary simplifications. The calculated results are taken statistically in terms of the district of meteorological stations. The results show that the peak values of the profiles at upper and lower parts in convective clouds are all lower (the peak values δ＝0.65~0.80) than the originals in MM4 model, and that the vertical distribution of heating rate on the region south of the Yellow River is rather uniform, but on the region north of the Yellow River the heating is mostly concentrated on some layer (the peak values δ＝0.50~0.65). All these are related to the climate condition and local circulation in each area of China. Substituting these vertical profiles of convective heating for the original heating parameters in MM4 model, it is found that the forecast accuracy on each precipitation level is improved greatly and the false alarm ratio is decreased during the last years of operational application