Abstract: Accurately predicting the probability of slope failure under rainfall is an important prerequisite for preventing rainfall-induced landslide disasters. Due to geological deposition and post-deposition, soil hydraulic parameters (i.e., saturated permeability coefficient) and shear strength parameters (i.e., effective cohesion and effective friction angle) of slopes exhibit spatial variability. In addition, the calculated probability of slope failure under rainfall infiltration considering the multi-parameter spatial variability is often larger than the observed frequency of slope instability. Thusly, an efficient approach is proposed for sequential probabilistic back analyses of multiple soil parameters of a slope under rainfall infiltration by successively integrating the field observations, i.e., the slope keeping stable before rainfall, the slope keeping stable after 57 days of weak rainfall, and the slope becoming instable after 3 continuous days of heavy rainfall. The proposed method is adopted to conduct the probabilistic back analysis of multiple spatially variable soil parameters for a two-dimensional unsaturated slope to update the estimation on the uncertainties of soil parameters. Then the probability of slope failure under target rainfall infiltration is predicted. The results indicate that the probabilistic back analysis by the fusion of multiple field observations can effectively reduce the estimation on the uncertainties of soil parameters and obtain the probability distributions of soil parameters that are more consistent with the actual observations, based on this, the predicted probability of slope failure is more reasonable. The more observations are integrated, the greater the degree of uncertainty reduction of soil parameters is. The research outcomes can provide a theoretical reference for the risk assessment and prevention of rainfall-induced landslide hazards in rainy mountainous areas.