Abstract: The Monte Carlo simulations based on the random finite element method (RFEM) are conducted to assess the reliability of the James Bay dike. The RFEM method leads to predictions of reliability (or probability of failure) as opposed to the more traditional "factor of safety" measures of slope safety. Different independent random fields are generated for different soil properties and different soil layers and mapped onto the Gaussian points in the finite elements. Both the isotropic and the anisotropic spatial variabilities are considered, and the influences of various model parameters used in 1D and 2D random fields are investigated and compared. Five different models are used to show the effect of correlation structure on slope reliability, and the importance of selection of a proper model as well as its parameters is highlighted. A reliability-based characteristic design factor is defined based on the statistical analysis results. It is shown that only 10%～40% of the slopes, depending on the RFEM model, can reach the expected mean performance as one would expect in a deterministic analysis. It is thought to be due to RFEM taking into account the spatial correlation in a rigorous way, allowing slope failure to occur naturally along the path of the least resistance or the weakest points. The results show that the horizontal scale of fluctuation (SOF) has a dominant effect in the present case and that a model using an equivalent isotropic SOF, based on a simple equation without considering the geometry of the failure surfaces, may possibly result in an erroneous assessment of reliability.