This paper describes a distributed modeling approach based on improved TOPOMODEL applied for simulation of rainfall runoff to a 10039 km2 Yingluoxia watershed on the middle range of the Qilian Mountains in Heihe River Basin,Northwest China.The simulation processes mainly assembled a modeling system centered on TOPMODEL,but the improvement was made on 1) the input parameter,precipitation,was given in a distributed mode through spatially interpolation from seven rain gauges to account for the spatial variations of precipitation due to strong or graphic affection in mountainous watershed by means of PRISM approach in consideration of topography,slopes,facets,distances to the meteorological stations as well as land use/covers; 2) simulation of saturated excess runoff was mainly based upon topography,but slightly modified following Campling et al.,(2002) to allow for the fact that in the upland sections of the hill slope the water table is not parallel to the surface topography by introducing a reference topographic index into the local storage deficit calculation,the other components deemed relevant to the hydrologic processes in the basin,however,were simulated not only upon the topography but also the Geographic Information System (GIS) data.Procedures were developed to generate grid model input files of basic meteorological components based on Digital Elevation Model (DEM) and land resource inventory GIS data.Model elements were 15 sub-watersheds delineated upon the channel network generated from DEM and a specified stream order threshold.Model element parameters are linked to GIS information averaged over each sub-watershed.Reference evapotranspiration was estimated following Penman-Monteith formula; a kinematic wave channel routing algorithm was applied on each sub-watershed to generate outlet total stream-flow.Totally 7 years of data series available for simulation study were broken in two parts and first part i.e.1990-93 was used for model parameter calibration and remaining data series i.e.1994-1999 was used for model validation.Simulation results are encouraging.Model efficiency was more than 0.82 both for model calibration and validation on independent data series.Stream-flow simulation are sensitive to uncertainty in the precipitation due to its significant spatial variability,and this uncertainty takes predominate role among all the possible sources such as air temperature and albedo…etc.