Abstract: There is extremely fragile ecosystem in the Tibetan Plateau in China because of its severely natural conditions. Once destroyed, it is very hard to get recovered. Especially it is main headwaters for famous rivers such as the Yangtze River, the Yellow River, etc. However, in recent years, China government has been conducting great road construction in the area. These road constructions would cause erosion more and more serious, and sediment from road would destroy aquatic ecology. Now it is necessary to find road erosion principles so as to take scientific and effective measures to reduce or control road erosion adverse effects on aquatic ecology, so we took two years to observe by setting up natural runoff plots on road sideslope (fill slope) of Qinghai-Tibet highway. When each rain was over, we sampled and attained runoff depth and soil loss. Siphonal Automated Record Rain Gauge recorded the rainfall process. After analyzing these data, we get the following results: (1) The product of rainfall and average rainfall intensity (PI) is well correlated to both runoff depth and sediment yield per rainfall event. In addition, the correlation coefficient between runoff depth and sediment yield per event also can get to 0.771. (2) Runoff depth and sediment yield decrease as slope length increases. (3) Runoff and sediment decrease over time, and the disturbance to plots got recovered during the past two years, so time is a very important factor to affect road erosion. (4) According to the existing datasets, soil losses of road sideslope is 119.91 t /(ha穉), including rainfall erosion and freezing and thawing erosion. The results of this study show us that the idea of USLE (Universal Soil Loss Equation) in America can be used for road erosion prediction after revised under specific conditions. Overall, this study explored the factors influencing road erosion and how they affect, so it is significant not only to make road erosion prediction model, but also to protect environment especially for watershed ecology.