Abstract: Abstract: The accurate simulation or prediction of precipitation runoff has been considered as one of the most important bases for resource management and environmental quality assessment of water and soil. The soil conservation service-curve number (SCS) model, one of the most popular runoff prediction models, cannot effectively determine the effect of the antecedent runoff condition (ARC) on runoff amount, which limits the accuracy of the model's runoff prediction. Assuming that antecedent daily precipitation depleted by evapotranspiration and seepage was linear with watershed water storage amount, the new ARC was established based on the recurrence relation of daily rainfall amount and watershed maximum rainfall storage amount. The SCS model was improved by correlating the initial abstraction with the new parameters of the potential initial abstraction and effective rainfall influence coefficient. The potential initial abstraction determines the maximum watershed rainfall storage amount prior to runoff and the threshold of daily effective rainfall amount, and the effective rainfall influence coefficient describes the dynamic depletion of antecedent daily effective rainfall amount induced by evapotranspiration and seepage. To reduce the number of unknown parameters, the relationship between the potential initial abstraction and the curve number was established under the condition that there was no rainfall for a long time prior to runoff. The data of precipitation and runoff amount from 1997 to 2008 required to assess the original and improved SCS models were collected from the 3 drainage areas of 1600 m2, 0.06 km2 and 1.36 km2 in the northern part of the Huaihe River basin, China. As antecedent daily precipitation period was 5 d and initial abstraction coefficient equaled to 0.2, the least-squares estimation method was used to calibrate the model parameters, i.e. the effective rainfall influence coefficient and the curve number, and the percent bias (PBIAS), Nash-Sutcliffe efficiency (NSE) and coefficient of determination (R2) were utilized to compare and assess the performance of the original and improved SCS models. The improved SCS model predicted daily runoff amount more accurately than the original model, and the improved SCS model increased the R2 and NSE by 27.0%-30.9% and 1.0%-78.3%, respectively, compared with the original during the validation period. Both models were calibrated with the close curve number, the effective rainfall influence coefficient was relatively stable, and the coefficient variation of 25% at plot scale resulted in the runoff prediction variation of less than 5%. The improved SCS model would perform better if it is applied in the areas with high evapotranspiration and seepage.