Abstract: Meso-scale atmospheric models have been extensively used in numerical weather prediction, regional climate research, and disastrous weather research. With the expansion of the applied area of atmospheric models, the higher and higher degree of precision is required. To meet the requirement, much research work has been done, such as 4-D variation assimilation, assimilation of weather satellite data and meteorological radar data, and so on. However, while paying respect to those new techniques, the model runners often ignore adjusting initial model parameters of topography, agrology, and hydrology, which might exert a crucial influence to the numerical simulation results. To demonstrate the impact of those initial model parameters, and display how to use meso-scale atmospheric models correctly, we simulated the weather of HEIFE region using RAMS with the initial surface characteristic data from USGS (U.S. Geological Survey), and adjusted only initial model parameters of topography, agrology, and hydrology and the related parameterization schemes. Based on the results of numerical simulations of the radiation balance and ground layer air temperature in HEIFE region during the periods from 20 June, 1991 to 20 July, 1991 with more than 40 different sets of initial parameters, the applicability of the regional atmospheric model system, RAMS, in northwest China, which is characterized by complicated topography and diversiform underlying surface, is demonstrated. The results show that the RAMS has moderate capability in numerical simulations of radiation fluxes and ground layer air temperature in HEIFE region even if given only the standard input data (the NCEP gridded reanalysis data file, upper air observations file, and surface observations file) required, but the errors of the results under this condition might be apparent in some cases. In fact, we found that when compared with observation data, intolerable errors always occurred if we concerned the results of upward long wave radiation and 2 m air temperature, no matter which kind of radiation transfer parameterization scheme and land surface parameterization scheme were used. Only based on the sound initialization of initial model parameters, especially initial soil temperature and soil moisture content, can the simulation results of RAMS be matched up well with the actual measurements. If higher degree of accuracy of the simulation results is desired in case of complicated weather conditions, the soil thickness would be another very important model parameter, and we should initialize it with a value no less than 4 meters. And soil moisture content should be initialized vertically nonhomogeneous with authentic data. The results of our research should be useful in improving the quality of simulation when applied to RAMS.