Abstract: A mesoscale experimental forecast system has been established in the CRAY-C92 computer in National Meteorological Center (NMC) based on PSU/NCAR’s mesoscale model MM5 (version 2), to which the local real-time data resources are linked. In order to investigate the feasibility and the accuracy of the mesoscale system for real-time prediction service under the current conditions of computer resources and of the available real-time data in NMC, the real-time forecast experiments are conducted for one and half months during the flood season of northern China in 1996 and in the special weather service period of “Hongkong returning to China” in 1997 using the system with 30 km horizontal resolution. The real-time forecast and verification results show that the forecasts of the precipitation amounts and the heavy rain locations are improved by MM5 more obviously than those by operational model (HLAFS): The mesoscale system could be used as a supplement tool for the weather forecast of city and precipitation model forecasts in higher frequency (less than 6 hours) could not be successful under the current data and computer resources.