Vegetation restoration in loess hilly areas effectively controls soil erosion. However, the impact mechanisms of vegetation canopy, litter, and roots on soil erosion under complex rainfall regimes remain uncertain. The typical sub-shrub species (Artemisia sacrorum) set up four treatments (natural condition, NC; canopy + roots, CR; roots only, OR; and bare soil, BS) in 2015-2017. K value clustering analysis was used to classify three rainfall regimes: Regime-I (short duration, small amount and medium intensity), Regime-II (medium duration, amount and high intensity), and Regime-III (long duration, large amount and low intensity). The results showed that the average runoff, soil loss and sediment concentration followed the order: BS treatment > OR treatment > CR treatment > NC treatment. The canopy accounted for more than 60% of the total contribution to soil loss and sediment concentration reduction. The litter had the lowest relative contribution to both runoff and soil loss reduction (29.76% and 4.19%, respectively). Roots accounted for the most contribution of 37.27% to reduce runoff, and just contributed 28.41% to reduce soil loss. Rainfall regimes significantly influenced the contribution of Artemisia sacrorum components to control soil erosion. Regime-III showed the highest contribution of different treatments to runoff and soil loss reduction (51.5%-79.9% and 65.7%-99.1%, respectively), while the lowest contributions occurred in Regime-II (14.4%-47.5%) and Regime-I (6.89%-95.08%), respectively. The canopy accounted for the 180.02% of the total contribution to reduce soil loss in Regime-II, but it instead had the minimum relative contribution of 6.81% in Regime-III. The relative contribution of roots to runoff reduction reached 91.24% in Regime-III, but just was 37.31% in Regime-II. The litter and roots even had negatively relative contribution in Regime-II or Regime-III. Besides, contributions of canopy and roots showed increase and decrease responses to increasing I30, respectively, while the litter contribution only negatively related to rainfall duration. These findings indicated that the canopy was the most critical factor in controlling soil erosion, while the litter played the lowest function, and the roots had continuous effects. The functions of plant components would change with different regimes, which maybe not always beneficial for reducing soil erosion. This study provides new insights into the rainfall-vegetation-erosion relationship.