In recent years, with the rapid development and application of GIS in China, the distributed hydrological model closely integrated with GIS has become a focus and hotspot of hydrological forecasting research. To build a more accurately distributed hydrological forecasting system, one of its core problems is to extract water flow direction matrix in a river basin based on DEM. However, karst regions in south China possess a binary spatial structure of both surface and underground, where there are a lot of funnels, depressions and caverns developing on the surface. The conventional method of filling depression will change the flow direction of true grid cells in the karst regions and result in a large number of flat areas. Furthermore, there are a large number of parallel water systems and channels to be lost in the extracted water system. This thesis can provide extracting a real water system under the background of karst topography by using digital river network technology to amend DEM without filling depression. There are almost no parallel water systems and channels to be lost in the extracted river. The water system extracted with this method has a better actual value to establishing a more precisely distributed hydrological model in karst area.