In the Soil Conservation Service Curve Number (SCS-CN) method for estimating runoff, three antecedent moisture condition (AMC) levels produce a discrete relation between the CN and soil water content, which results in corresponding sudden jumps in estimated runoff. Singh et al. (2015) present an improved soil moisture accounting (SMA)-based SCS-CN method that incorporates a continuous function for the AMC which can obviate sudden jumps in estimated runoff. However, their method ignores the effect of storm duration on surface runoff, yet this is an important component of rainfall–runoff processes. In this study, the Singh et al. (2015) method for runoff estimation was modified by incorporating storm duration and a revised SMA procedure. Then, the performance of the proposed method was compared to both the original SCS-CN and Singh et al. (2015) methods by applying them in three experimental watersheds located on the Loess Plateau, China. The results indicate that the SCS-CN method underestimates large runoff events and overestimates small runoff events, yielding an efficiency of 0.626 in calibration and 0.051 in validation; the Singh et al. (2015) method has improved runoff estimation in both calibration (efficiency = 0.702) and validation (efficiency = 0.481). However, the proposed method performed significantly better than both, yielding model efficiencies of 0.810 and 0.779 in calibration and validation, respectively.