In current slope reliability analysis, the failure probability may be wrongly calculated because of the inadequate consideration of anisotropic spatial variability of soil parameters. Therefore, a random finite difference method (RFDM) framework considering anisotropic spatial variability is established and taking general anisotropic spatial variability and a cohesive-frictional soil slope as an illustrative example. The results show that the coordinate-transformation-based anisotropic random field discretization method can effectively characterize anisotropic spatial variability of soil parameters; Strain-clustering-based slope critical slip surface searching algorithm can accurately determine the complex critical sliding surface of slope; Compared with the general anisotropic spatial variability, the slope failure probability is overestimated and greatly underestimated when considering rotational anisotropy and transverse anisotropy, respectively; In addition, considering isotropic random fields can overestimate and underestimate the slope failure probability in case of larger and smaller scale of fluctuation, respectively.