Precipitation is an essential input parameter for model research of hydrology, ecology and climate, and accurate precipitation data with high resolution are very important for the understanding of regional climate change. At present, TRMM satellite with resolution of 0.25 degree can provide global precipitation data, but the application of the lower resolution of TRMM 3B43 precipitation data are restricted in hydrological or climatic model of watershed scale. In mountain, precipitation data cannot be obtained by simple interpolation or extrapolation, so it is necessary to apply downscaling methodology to improve the resolution of TRMM precipitation data. In this article, the middle part of Tianshan Mountains is study area, based on the correlation of annual precipitation with the regression model between the TRMM 3B43 annual accumulated precipitation data and such factors as DEM, NDVI, aspect, longitude and latitude is built. TsHARP is a statistical downscaling algorithm, which is improved by introducing DEM and local Moran's I index of NDVI, and the precipitation data with 250 m resolution in the studied area during 2001-2006 are obtained. From the results, it can be known that the internal precipitation in study area affected by the altitude is serious. The amount of precipitation in the western Tianshan Mountains is more than the eastern and the amount of precipitation in the northern slope is more than the southern slope. This conclusion and the previous research on the precipitation are consistent. The downscaled results are validated with observed precipitation data from meteorological stations in the studied area. It is obtained that 1) the errors of downscaled results are substantially less than those of original TRMM precipitation data, and the maximal improved value of annual average precipitation is 70mm; 2) the spatial distribution characteristics of precipitation in the middle section of Tianshan Mountain can be depicted in detail by downscaled results; 3) the correlation coefficient (R~2) for TRMM 3B43 accumulated precipitation data and observed precipitation data is 0.618, while the one for downscaled results and observed precipitation data is 0.713. So the downscaling methodology built in this article is feasible, in a certain extent, which can provide technical support for the research of rigion with high altitude where human beings cannot monitor the precipitation area, and can provide spatial distribution of precipitation with detailed description of the results in mountains.