[Objective] Under natural rainfall conditions, a large amount of soil nutrients are losing with overland flow or surface runoff on sloping farmlands in the loess area, which exacerbates the decline of soil quality and productivity of the farmlands and causes serious agricultural non-point source pollution to the environment. Therefore, it is particularly important to accurately predict how soil nutrients loing with surface runoff in the loess areas. However, the existing nutrient loss prediction model focuses on estimating the total nutrient loss over a long period of time in an area, but neglects the effect of infiltrating water diluting the nutrients in the exchange layer before runoff starts.[Method] In this study, according to the characteristics of nutrient migration in different time periods of a rainfall event, the entire process of a rainfall event is divided into three phases:(1) from the beginning of rainfall (t₀) to the time when the exchange layer is completely saturated (t_sa), (2) from the time when the exchange layer is completely saturated (t_sa) to the time when runoff occurs (t_p), and (3) from first runoff (t_p) to the end of the rainfall. Based on the approximate analytical solution of the motion wave model, an approximate analytical model of nutrient loss was established, which integrated the effects of raindrop splashing, diffusion and infiltration;and then a simulated rainfall experiment was conducted to determine parameters of the model and verify reliability of the model.[Result] Results show the values of surface runoff and nutrient loss predicted with the model accurately matched the measured values (R²> 0.8, Nash-Sutcliffe efficiency coefficient> 0.347). In this nutrient loss model, the raindrop-induced water transfer rate e_r was valued in the range of 0.006~0.023 cm·min^-1, and the exchange layer depth d_e in the range of 0.68~1.32 cm. The former significantly affected peak rate of the nitrate nitrogen and ammonium nitrogen loss, whereas the latter did range of the overall variation of the loss rate and boosted total loss of the nutrients. The model was found to be more sensitive to d_e than to e_r.[Conclusion] Some field management measures, such as revegetation, deep fertilization, etc. should be adopted to reduce raindrop kinetic energy and lower the content of nutrients in the exchange layer, so as to realize the purpose of reducing nutrient loss. In general, this approximate semi-analytical model has fully considered the influence of water infiltration in unsaturated soil on nutrient transport in the exchange layer, so it can be used to predict nutrient loss processes on bare slopes in arid and semi-arid regions. However, it should be emphasized that accurate calculation of runoff process is the basis of the nutrient loss simulation, so it is advisable to select an appropriate infiltration formula and solute adsorption coefficients in the light of soil texture, nutrient type, and nutrient concentration in rainwater.