Field rainfall simulation experiments at two sites are used to partition sediment transport on unpaved roads into splash and hydraulic erosion components. Rain splash processes contributed 38--45% of total sediment output, with instantaneous contributions being variable throughout 60-min high-energy events. For low- and medium-magnitude rainstorms, splash erosion on roads is initially controlled by the removal of easily erodible material, followed by a dramatic reduction in sediment output associated with limited detachment from the resistant, highly compacted road surface. A conceptual model explaining temporal variations in splash and hydraulic erosion as functions of prestorm surface preparation (via traffic, maintenance, and mass wasting processes) is presented. For situations where loose sediment is readily available, rain splash energy is less important to sediment detachment. If the loose layer is diminished (e.g., following an overland flow event) or protected by a surface crust, splash energy is needed to detach material from the road surface. Equations in most physically based erosion models do not predict temporal variations in road sediment transport that result from the removal of a loose surface layer of finite depth. A strategy that successfully treats this removal as changes in road erodibility is introduced. ¿ 2000 American Geophysical Union