Abstract: In general, limited test data can be collected from geotechnical site investigation. However, it is typically difficult to accurately characterize the spatial variation in soil properties with limited test data. This paper aims to propose a probabilistic back analysis and reliability updating approach considering the spatial variability of soil properties. With this approach, multiple sources of test data including laboratory and in situ test data can be utilized to rationally back analyze the spatially varying soil properties and update the slope reliability. The implementation procedures of the proposed approach are presented step by step. In addition, a non-stationary random field model of undrained shear strength is developed for proper characterization of the prior information of soil property. Finally, a clay slope under undrained conditions is investigated to demonstrate the effectiveness of the proposed approach. The influences of the test data and borehole location on the posterior probability of slope failure are also addressed. The results indicate that the proposed approach can effectively back analyze the spatially varying soil properties and update the slope reliability. By incorporating multiple sources of test data into the Bayesian analysis, the estimated means of soil parameters are consistent with the test data. The uncertainties of soil parameters are greatly reduced and the slope reliability is significantly improved. Due to spatial variation, test data has a stronger effect on the updating of soil parameter statistics with short distances to the borehole locations of measurement, compared with soil parameter statistics with long distances to the borehole locations.