High-resolution precipitation distributions in mountainous areas are important for hydrological and ecological assessments, especially in regions with few weather stations. In this study, we proposed an improved model for precipitation downscaling by adding two new parameters, i.e. the maximum precipitation increment direction and the prevailing precipitation direction, which represent the impacts of elevation and the sources of precipitation, respectively. The model parameterization is based on observations made at meteorological stations, terrain factors (e.g. elevation, aspect, and slope), and the new parameters. To evaluate the model, we used six sub-models, each of which considers different influencing factors, to estimate the precipitation distribution and compare their estimation errors. Based on the mean absolute error (MAE) and the root-mean-square error (RMSE) at the validation stations, we found that the sixth sub-model, which includes all the influencing factors, clearly ranks above the others in terms of precipitation downscaling. The monthly MAE and the RMSE of our downscaled precipitation range from 2.2 to 16.1 mm and from 3.4 to 22.7 mm, respectively, indicating more accurate estimation than the raw tropical precipitation measuring mission (TRMM) products (monthly MAE: 3.6-22.0 mm; monthly RMSE: 5.1-28.6%). Our results also show that the sixth sub-model performs better than the Auto-Searched Orographic and Atmospheric Effects Detrended Kriging model (ASOADeK model or Guan's model) due to the inclusion of the elevation and the sources of precipitation. Based on the sixth sub-model and the TRMM 3B43 products, we developed the monthly precipitation products in China from 2000 to 2007 at a spatial resolution of 1 km. Our improved approach to precipitation downscaling could be used for regions where the precipitation distribution is greatly affected by the terrain and few observations are available for estimating the precipitation distribution.