Soil detachment is an unavoidable path and important process of slope soil erosion and sediment yield. Precise prediction of the process of soil detachment is of great significance to the consummation of a physical model for soil erosion. A runoff scouring experiment, designed to have a hydraulic flume, of which the slope gradient is adjustable with the range of 8.8% ~ 46.6%, and various flows (0.5 ~2.5 L s^-1) was carried out to study relationships of soil detachment rate with slope, flow, shear stress, stream power and unit stream power on yellow soil slopes in Southwest regions, and difference between yellow soil and loess in soil detachment rate in the case of the same flow rate and typical slope of the regions. Results show that the soil detachment rate of yellow soil in the Southwest region increases with the flow and slope. Multiple regression analysis of the slopes and flows may be used to predict detachment rate of the soil very well (R²=0.9). Stream power and unit stream power both display good power function relationships with soil detachment rate of the yellow soil, with determination coefficient(R²) being 0.83 and 0.79, respectively. However, flow shear stress is not a good factor to predict soil detachment rate (R²=0.18). In Loess, the soil detachment rate is obviously higher than in yellow soil, and the difference between the two soils in soil detachment rate widens with increasing slope scouring flow. Although flow, slope, stream power and unit stream power all can be used to predict soil detachment rate, and their regression equation differ slightly from those in other domestic researches, the coefficients that characterize soil erodibility in the equations differ sharply, which reflects the particularity of the soil erosion process of yellow soil being affected by runoff.