Through in situ observation experiment in the field micro-plots, the 34 erosive rainfall patterns from 2017—2021 were classified based on the concentrated periods of rainfall during the rainfall process, and the effects of rainfall pattern variation on the interrill erosion on loess slopes and slope effects were systematically analyzed. The results showed that the advanced rainfall pattern was dominant, accounting for 52% of the total rainfall, while the intermediate rainfall pattern and delayed rainfall pattern each accounted for 24%. The contribution of intermediate rainfall pattern to total runoff yield (25.56%±3.83%) was significantly lower than that of the advanced rainfall pattern (36.70%±2.07%) and delayed rainfall pattern (37.74%±4.01%) (p<0.05). The average runoff yield of different rainfall patterns was in the order of delayed rainfall pattern > intermediate rainfall pattern > the advanced rainfall pattern, and the delayed rainfall pattern and intermediate rainfall pattern was significantly higher than that of advanced rainfall pattern. The average sediment yield of different rainfall patterns was in the order of delayed rainfall pattern > advanced rainfall pattern > intermediate rainfall pattern. The rainfall of the delayed rainfall pattern was significantly higher than that of the intermediate rainfall pattern (p<0.05). The variation of slope effect of rainfall pattern on the effect of interrill erosion is not obvious, but with the increase of slope, the average runoff and sediment yield of delayed rainfall pattern were larger than that of advanced and intermediate rainfall pattern, indicating that the sediment yield of delayed rainfall pattern was relatively sensitive to the variation of slope. In the nonlinear fitting models of runoff, slope and rainfall parameter combination and sediment yield, the equation decision coefficients were delayed rainfall pattern (0.939~0.942) > the intermediate rainfall pattern (0.776~0.845) > the advanced rainfall pattern (0.220~0.537), indicating that the classification of rainfall pattern could improve model simulation accuracy.