Infiltration of rainfall can lead to possible slope failures. The hydraulic properties exhibit inherent spatial variability and hence it can impact the rainfall infiltration in reality. However, the spatial variability is difficult to be characterized when data are limited. An efficient method for characterizing the spatial variability of saturated hydraulic conductivity based on monitored pore water pressure responses of a soil slope is proposed. A hypothetical random heterogeneous slope subject to steady-state rainfall event is modeled using the Karhunen–Loève expansion method and is estimated with Markov Chain Monte Carlo method. To reduce the computation load of probabilistic back analysis, the polynomial chaos expansion model is used as a surrogate model to approximate the finite element model. Results show that the inverse procedure has a good performance and the estimated field has high precision. The uncertainty of the estimated ks distribution is significantly reduced and the estimated field can represent the true field mostly.