Accurate prediction of soil detachment capacity is fundamental to establish processbased erosion models and improve soil loss assessment. Few studies were conducted to reveal the mechanism of detachment process for yellow soil on steep cropland in the subtropical region of China using field experiments. This study was performed to determine soil detachment characteristics and explore the relationships between soil detachment capacity (D c ) and flow rate, slope gradient, mean velocity, shear stress, stream power and unit stream power. Field experiments were conducted on intact soil with flow rates ranging from 0.2 × 10 −3 to 0.5 × 10 −3 m −3 s −1 and slope gradients varying from 8.8 to 42.4%. The results showed the following. (a) D c of yellow soil was smaller than other soils because of its high clay content. (b) D c was more susceptible to flow than to slope gradient. Power functions were derived to depict the relationship between D c and the flow rate and slope gradient (R 2 = 0.91). (c) D c was better simulated by power functions of the stream power (R 2 = 0.83) than functions of the shear stress or the unit stream power. (d) Considering its accuracy, simplicity and accessibility, the power function based on flow rate and slope gradient is recommended to predict D c of yellow soil in the field. The results of this study provide useful support for revealing soil detachment mechanism and developing processbased soil erosion models for the subtropical region of China. K E Y W O R D S detachment capacity, erosion prediction, hydraulic parameters, karst regions, soil erosion, steep slopes Over the past several decades, two main experimental procedures, filed plot experiment and laboratory flume experiment, were generally used to determine soil detachment rate or capacity (Knapen, Poesen, Govers, Gyssels, & Nachtergaele, 2007). Laboratory flume experiments were widely used to understand fundamental erosion processes. Small sample area and flexible setups allow laboratory tests easy to control and replicate. Numerous results from laboratory flume experiments have been reported to relate the D r or D c to hydraulic parameters, such as the flow regime, flow rate, slope gradient, flow depth, velocity, shear stress, stream power and unit stream power, under different conditions. Lyle and Smerdon (1965) conducted flume experiments and indicated a linear relationship between soil erosion and shear stress for Texas soils, but the conclusion was obtained from a constant gentle slope (0.2%). Subsequently, Nearing et al. (1991) used a hydraulic flume to study the process of soil detachment. The results showed a logarithmic relationship between detachment capacity and flow depth, slope gradient, and mean weight diameter of the aggregates. The results also indicated that soil detachment was greatly influenced by flow depth while weakly affected by slope. Nearing's experiments were well-conducted, but the test slopes were very low, ranging only from 0.5 to 2.0%. Due to the fact that much of the cultivated farmland in China lies ...