Slope gradient is an important factor affecting raindrop splash erosion. Hence its effect was investigated using soil trays simulating nine slope gradients (9%, 18%, 27%, 36%, 47%, 58%, 70%, 84%, and 100%), and each soil tray was 0.5m × 0.5m × 0.1m (horizontal length × width × depth) in dimension, containing loamy soil, collected from Yanqing, Beijing. All the treatments were subjected to a simulated rainfall with a constant intensity of 67 mm h^-1 for 1 hour. Specially designed samplers were used to partition total splash into directional components (e.g. upslope, downslope, and lateral). Results indicate that splash erosion occurred in the following three patterns: (1) the rate was low prior to the generation of runoff, peaked rapidly at the beginning of runoff, decreased gradually and leveled off; (2) the rate was high prior to the generation of runoff and dropped rapidly after runoff was generated and then leveled off at a certain rate; or (3) the rate did not show any obvious fluctuation during the rainfall event. In all the treatments, the upward raindrop splash erosion rate was always lowest and the downward one was the highest. And the downward splash and sideward splash were negatively related to slope gradient and the relationship could be fit with a linear or exponential function. The downward splash erosion varied more obviously than lateral and upslope splash erosion, which indicates that downslope splash erosion is more significantly affected by slope gradient. The downward splash erosion, total splash detachment, and net splash transport increased initially with rising slope, and then decreased at the critical slope gradient, 35° (70%), all showing linear relationships.