The combined use of water erosion models and geographic information systems has facilitated soil loss estimation at the watershed scale. Tools such as the Geo-spatial interface for the Water Erosion Prediction Project (GeoWEPP) model provide a convenient spatially distributed soil loss estimate but require discretization to identify hillslopes and channels. In GeoWEPP, the TOpographic PArameteriZation (TOPAZ) model is used as an automated procedure to extract a watershed boundary, hillslopes and channels from a digital elevation model (DEM). Previous studies in small watersheds have shown that the size of the hillslopes and the channel distribution affect the model estimates, but in large watersheds, the effects on the soil loss estimates have yet to be tested. Therefore, the objective of this study was to evaluate the effect of discretization on the hillslope sediment yield estimates using GeoWEPP in two large watersheds (>10km(2)). The watersheds were selected and discretized varying the TOPAZ parameters [critical source area (CSA) and minimum source channel length (MSCL)] in a 30-m resolution digital elevation model. The drainage networks built with TOPAZ were compared with each other using the drainage density index. The results showed that the discretization affected hillslope sediment yield estimates and their spatial distribution more than the total runoff. The drainage density index and the hillslope sediment yield were proportional but inversely related; thus, soil loss estimates were highly affected by the spatial discretization. As a result of this analysis, a method to choose the CSA and MSCL values that generates the greatest fraction of hillslopes having profile lengths less than 200m was developed. This slope length condition is particularly crucial when using the WEPP and GeoWEPP models, in order for them to produce realistic estimates of sheet and rill erosion. Finally, and as a result of this analysis, a more reliable method was developed for selecting the TOPAZ channel network parameters (CSA and MSCL). Copyright (c) 2015 John Wiley & Sons, Ltd.