Abstract: Abstract: Overland runoff is a main factor causing soil erosion ion hillslope, and runoff modeling is an effective way to predict soil erosion for conservation of the hillslope cropland of purple soil. The model of Soil Conservation Service-curve number (SCS-CN) is widely used for predicting direct runoff discharge based on measured rainfall, and also used as a runoff-estimating component of more complex watershed models. As a basic input parameter in the SCS-CN model, the curve number greatly affects the predictive value of runoff depth, and the initial abstraction ratio (λ) that was assumed to be o 0.2 in the original development of SCS-CN model. When applied in the hillslope areas, few attempts have been made to incorporate a slope gradient factor into the CN method although slope gradient exerts great influence on discharge of overland runoff according to relevant studies. In addition, the initial abstraction ratio is considered to be a constant in many applications, although lots of studies revealed that variations of the ratio exist in different conditions. In this study, the relationship between runoff and slope gradient was revealed by analyzing observed data of 5 rainfall-runoff events in 2013 from experimental plots with slopes varying from 6.5° to 25°. The existing approaches integrating slope gradient factor in the SCS-CN model were directly evaluated for simulating the rainfall-induced runoff in the hillslope cropland of purple soil. Meanwhile, the initial abstraction ratio was optimized considering the influence of rainfall volume. The Nash-Sutcliffe efficiency (E) and relative error (RE) were used in the simulation results evaluation. The results indicated that the observed discharge of runoff increased obviously with the increase of slope gradient in a rainfall event, and the runoff discharge achieved maximum at a critical slope. For our experimental conditions, the slope-modified SCS-CN method proposed by Williams overestimated the runoff depths with Nash-Sutcliffe efficiency (E) of 0.76. The initial abstraction ratio (λ) of 0.2 was applicable to estimate small rainfall-runoff events based on the equation proposed by Huang. For the large rainfall-runoff events, the slope-adjusted CN method put forward by Huang performed better with increasing initial abstraction ratio (λ). It appeared to be appropriate for overland runoff prediction in hillslope cropland of purple soil when the value of λ was equal to 0.3 in the case of large rainfall-runoff events, according to the values of Nash-Sutcliffe efficiency (E=0.99) and mean relative error (RE=7.42%). Runoff discharge observed and estimated by the modified model achieved relatively good agreement in each rainfall-runoff event with mean relative error of 7.58%, 7.93%, 8.92%, 5.98%, and 7.13%, respectively. The results also revealed that the validated curve numbers calculated from the equation developed by Huang were 78.23, 78.45, 78.77, 79.11, and 79.47 for each slope gradient from 6.5°to 25°, respectively. This study provides valuable information for rainfall-runoff estimation and soil and water conservation in the hilly area of purple soil.