The uncertainty of modeling input will increase the simulation error, and this situation always happens in a model without user-friendly interface. WinVAST model, developed by the University of Virginia in 2003, treats an entire multi-catchment by a tree-view structure. Its extra computer programs can connect geographic information system (GIS). Model users can prepare all the necessary information in ArcGIS. Extracting information from GIS interface can not only decrease the inconvenience of data input, but also lower the uncertainty due to data preparation. The Daiyuku Creek and Qupoliao Creek in the Fei-tsui reservoir watershed in Northern Taiwan provided the setting for the case study reported herein. The required information, including slope, stream length, subbasin area, soil type and land-use condition, for WinVAST model should be prepared in a Microsoft Access database, which is the project file of WinVAST with extension mdb. In ArcGIS interface, when the soil layer, land-use layer, and Digital Elevation Model (DEM) map are prepared, all the watershed information can be created as well. This study compared the simulation results from automatically generated input and manual input. The results show that the relative simulation error resulting from the rough process of data input can be around 30% in runoff simulation, and even reach 70% in non-point source pollution (NPSP) simulation. It could conclude that GIS technology is significant for predicting watershed responses by WinVAST model, because it can efficiently reduce the uncertainty induced by input errors.