Abstract: Abstract: Bench terraces have been playing an important role in the control of soil and water loss in the loess plateau of China. However, it is still lacking on the effect of terrace construction on the rainfall-runoff processes in a watershed. Meanwhile, the Soil Conservation Service Curve Number (SCS-CN) model also remains rarely used to predict the hydrological evolution in the watershed that is distributed with the wide terrain modification including bench terraces. Since 1995, large-scale bench terraces have been constructed in the Luoyugou watershed of Gansu Province (34°35′21.7″-34°39′22″N, 105°30′33.8″-105°44′2.5″E) in China, particularly in the loess hilly and gully region. Taking the Luoyugou watershed as the research object, this study aims to clarify the influence of the terrace on the rainfall-runoff, further calibrating and improving the SCS-CN model. The rainfall-runoff process data was firstly collected in the study area. The 88 rainfall-runoff events were then selected, where 34 events occurred in 1985-1990 (before the terrace construction) and 54 events in 2000-2010 (after terrace construction). The results showed that the performance of the terrace in controlling runoff processes depended mainly on the rainfall events. The terrace significantly decreased in the peak discharge and the total runoff volume by more than 55% under the relatively small rainfall events, while only a very limited role was played in the peak discharge and runoff under the high rainfall events. The ratio of runoff to rainfall in the watershed significantly decreased from 0.13 in 1985-1990 to 0.06 in 2000-2010. All rainfall events were divided into small and moderate-heavy precipitation intensity events, according to the relationship between the rainfall and runoff. The least square method was first used to determine the parameters of the standard SCS-CN and two revised models by using the antecedent soil moisture. Then, an improved SCS-CN model was established, considering the rainfall amount and intensity. A correlation analysis was also made between the rainfall elements (rainfall, average rainfall intensity), and the event runoff curve numbers that were calculated from the collected rainfall-runoff data. The rainfall amount and intensity were then selected as the important factors affecting the event runoff curve numbers for the two types of precipitation. The improved model was finally verified to compare with the standard SCS-CN and two improved models using the antecedent soil moisture. The average initial loss rate in the precipitation was also calculated to verify the model. The results showed that the improved SCS-CN model was achieved in the best performance to effectively predict the rainfall-runoff process in the watershed distributed with bench terraces. Among them, the Nash efficiency coefficients were 0.291 and 0.439 for the small and high rainfall events, respectively, with the root mean square error (RMSE) of 1.0 m, and 1.88 mm, respectively. By contrast, the standard and revised SCS-CN model failed to simulate the rainfall-runoff process, although considering the antecedent soil moisture. The Nash efficiency coefficients of the improved SCS-CN model increased, whereas the RMSE were reduced, compared with the standard SCS-CN. The initial loss rates were 0.29 and 0.02 for the small and high rainfall events, respectively, which was fully consistent with the field observation. Anyway, the improved SCS-CN model considering the rainfall factors performed the best for the rainfall-runoff process in the watershed with terraces. This finding can also provide a strong important tool for the planning and management of soil and water conservation.