Abstract: Anisotropy exists in the spatial variability of soil parameters. Therefore, it is rational to indicate the spatial distribution of slope parameters using anisotropy random fields. Based on the anisotropic random field of shear strength of soil for a two-layer slope, the effects of vertical scales of fluctuation, horizontal scales of fluctuation and coefficient of variation (COV) of soil parameters on the slope failure probability, instability modes and risk assessments are studied. The main conclusions are drawn as follows: with the increase of COV, the risk of slope failure gradually increases. In low-variability soils, there is almost no risks of slope failure. On the whole, the failure probability of slope is consistent with the risk of failure as COV increases. The deep-layer slop mode accounts for a large proportion, but with the increase of COV, the deep-layer slope mode gradually becomes the shallow slope one. When the scale of fluctuation (including horizontal and vertical) increases, the failure probability of slope and risks increase accordingly. However, when the scale of fluctuation exceed a particular size, which is related to the size of the slope, the increasing amplitude of failure probability and risks slows down as the scale of fluctuation increases.