Combining a manually simulated rainfall experiment with nutrient loss model, we quantified and forecasted the relationship between nutrient loss of matrix runoff and rainfall frequency and time in a rock slope matrix-vegetation system. Results from the rainfall experiment indicated that the losses of nitrogen (N), phosphorus (P) and potassium (K) showed a negative relationship with rainfall frequency. The relationship between the accumulated nutrient loss and rainfall time followed the power function, and the nutrient loss showed significant differences with rainfall frequency and with rainfall time (P=0.00). These models could be used to forecast the nutrient loss accurately, with the coefficients of determination more than 0.86. The results of the model fitting agreed well with the experimental results, with the relative errors less than 27%. Loss of element K was best predicted in the first rainfall, while Losses of elements N and P were in the last rainfall. Therefore, our mathematical models could be used to describe the changing process of nutrient loss in runoff along with rainfall frequency and time, and provided a theoretical basis for nutrient management in the rock slope matrix-vegetation system.