Probabilistic back-analysis is an important means to infer the statistics of uncertain soil parameters, making the slope reliability assessment closer to the engineering reality. However, multi-source information (including monitored data, field observations and slope survival records) are rarely used in current probabilistic back-analysis. Conducting the probabilistic back-analysis of spatially varying soil parameters and slope reliability prediction under rainfalls by integrating the multi-source information is a challenging issue since thousands of random variables and high-dimensional likelihood function are usually involved. In this paper, a framework by integrating a modified Bayesian updating with subset simulation (mBUS) method with adaptive conditional sampling (aCS) algorithm is established for the probabilistic back-analysis of spatially varying soil parameters and slope reliability prediction. A highway slope case is investigated to illustrate the effectiveness of the established framework. Within this framework, the high-dimensional probabilistic back-analysis problem can be easily tackled, and the multi-source information can be fully used for the back-analysis. The results show that the obtained posterior knowledge of soil parameters is in good agreement with the field observations. Furthermore, the updated statistics of soil parameters can be utilized to predict the failure probability of a slope caused by the heavy rainfall event on September 12, 2004 in the district near the studied slope.