Soil erosion is a dynamic environmental process that influenced by multiple factors. However, most previous studies only examined the causative factors without ranking their relative importance or examining the individual factors. In this work, back-propagation (BP) neural network modelling and grey relational analysis were used to rank the effects of 7 factors—vegetation growth stage (VGS), vegetation type (VT), vegetation cover (VC), rainfall intensity (RI), rainfall duration (RD), antecedent soil moisture (ASM) and slope gradient (SG)—on total runoff (TR) and total sediment (TS) following simulated rainfall events at 5 intensities (30, 45, 60, 90, 120 mm h–1). The experimental plots including 4 treatments, bare soil (control), ryegrass (Lolium perenne L.), purple medic (Medicago sativa L.) and spring wheat (Triticum aestivum L.) under 4 different slopes (9%, 18%, 27.8%, 36.4%). BP models were constructed to predict TR and TS; their predictions tracked the experimental data very closely. A factor analysis based on the BP models ranked the influence of the 7 factors on TR and TS as RI > VC > ASM > RD > VGS > VT > SG and RI > VC > SG > ASM > RD > VGS > VT, respectively. Grey relational analysis provided similar results, ranking the effects of these factors on TR and TS in the order RI > VC > ASM > RD > SG > VGS > VT and RI > VC > SG > ASM > RD > VT > VGS, respectively. These results indicate that runoff and sediment yield depend most strongly on RI and VC, while the effects of the other factors are less pronounced.