Themodeling of soil erosion by water supposes an accurate and thorough understanding of the hydrology. Thus, it is critical to have a good delineation of the surface flow path. The flow network data are critical for important uses such as flood forecasting and watershed management. Geographic Information System (GIS) functions are able to compute a flow network directly from the digital elevation models. Because the flow directions are only based on the topography, the other factors controlling the flow directions are overlooked. In the agricultural areas, work such as tillage can have a large impact on the flow direction. We propose a 5-step procedure to account for such man-made features. The use of this procedure clearly improves the quality of the computed flow network. This procedure has been successfully implemented in a GIS and improves the prediction of surface flow and therefore improves water erosion modeling at the watershed scale.