Forest roads are a major source of and transport pathway for eroded sediments in mountainous watersheds. When rills develop on these roads' surfaces, they amplify sediment erosion. Best management practices can decrease sediment erosion, but in order to efficiently implement these practices it is necessary to determine which factors have the most influence on rill development on forest roads. Despite this need, there is scarce literature on rill development on forest roads. To fill this gap in knowledge, based on field survey and multivariate statistical methods including redundancy analysis (RDA) and variation partitioning analysis (VPA), we investigated unpaved forest roads in the Xiangchagou watershed in China and quantified the extent to which various factors influenced rill formation. Specifically, we studied how rill erosion intensity (REI) and rill morphological characteristics (like rill length, mean width and depth, density, and severity of fragmentation) varied along the slope of a forest road. We also introduced the concept of a road's hydrological constituents (its upslope catchment, surface, and cutslopes), and determined how much each constituent contributed to REI. We found that REI and morphological characteristics decreased moving from the upper portion of road segment downward, implying that rills developed more intensely uphill. Additionally, REI increased exponentially with rill width, density, and severity of fragmentation, and increase linearly with length and depth. Conversely, REI decreased exponentially with rill width-depth ratio. These relationships suggest that the morphological characteristics of rills could be used to predict the REI of a given road segment. Finally, we found that the road characteristics that best predicted rill formation included catchment area, cutslope area, and gravel bareness. Correspondingly, the upslope catchment, cutslopes, and road surface contributed 11.56%, 30.83%, and 8.23% of the variation in REI and morphological characteristics. The interaction between upslope catchment and road surface explained 19.89% of the variation. These results suggest that when best management practices are implemented to decrease erosion caused by forest roads in mountainous watersheds, they should integrate these hydrological constituents of a road.