In order to explore the effects of hydrodynamic erosion factors on soil detachment rate of freeze-thaw slope, two slopes (10°, and 15°), four flow rates (4.5, 6.5, 8.5 and 10.5 L/min), and four original thaw depth (2, 5, 10 and 15 cm) were designed to simulate filed runoff. By means of BP Neural Network and Stepwise Regression Analysis, we analyzed the relationship between the soil detachment rate and 7 hydrodynamic erosion factors, including stream power, unit stream power, original thaw depth, flow shear stress, flow rate, flow velocity and slope. The results showed that the rank of influence significance of hydrodynamic erosion factors from high to low was flow stream power＞unit stream power＞original thaw depth＞flow shear stress＞flow rate＞flow velocity＞gradient. Furthermore, according to BP Neural Network model, the average error of soil detachment rate prediction was 2.848% (R2=0.954). In view of Stepwise Regression Analysis model, the average error was 4.820% (R2=0.925). According to Single Hydrodynamic Erosion Factors (Stream power) model, the average error was 5.298% (R2=0.867). In general, the prediction effect based on BP Neural Network was better than others. This study provides a new idea for the establishment of soil erosion prediction model under different original thaw depths in the spring thaw period.