The impact of non-point source pollution on the water quality of the North Canal River is becoming increasingly prominent. In this study, the riparian buffer zones (RBZ) of the Nansha River and Beisha River, the inlet tributaries of the Shahe Reservoir in the North Canal basin, were selected to investigate the purification effect of riparian buffer zones on runoff pollution during the rainfall process. Two RBZ types, Type I RBZ (levee-flood control retaining wall-woodland-grassland) and Type Ⅱ RBZ (levee-woodland-grassland), were classified by the distribution characteristics of RBZ structure and plant communities in the North Canal River basin. The north bank of the Nansha River (NB) and the south bank of the Beisha River (BN) are typical of Type I RBZ, with low total vegetation cover, "short and steep" slopes, and low herbaceous cover but high diversity. The south bank of the Nansha River (NN) is a typical representative of Type Ⅱ RBZ, with "long and slow" slopes and high herbaceous cover (29.16%) but low diversity. In order to investigate the impacts of rainfall characteristics and RBZ types on the runoff pollutant, a 1 km area in each of the three RBZs was selected to carry out the RBZ non-point source pollution prevention and control engineering trials. The results indicated that Type I RBZ required less time and rainfall to produce runoff and had a greater peak runoff. Type Ⅱ RBZ produced runoff only under heavy rainstorm conditions, with greater runoff retention capacity. Energy dissipation ponds with gravel as the main fillers were set up at the runoff inlets of the RBZ, which effectively reduced runoff pollution. ρ(NH4+-N) and ρ(NO3--N) in the runoff were below 1.6 mg·L-1; ρ(TN) was below 5 mg·L-1; and ρ(PO43-P), ρ(DTP), and ρ(TP) were below 1.0 mg·L-1. The grass ditch of the RBZs effectively reduced ρ(NH4+-N) of the runoff. The retention rate of SS and the reduction effect of pollutants in Type Ⅱ RBZ were better than those in Type I except under heavy rainstorm conditions, which is related to the different RBZ structures and vegetation cover. The correlation analysis results showed that slope length, slope gradient, vegetation cover, and rainfall characteristics were significantly correlated with runoff SS, COD, nitrogen, and phosphorus pollution.