The response of the seepage field to the reservoir water level change greatly depends on the saturated hydraulic conductivity of reservoir landslides, which plays a critical role in their evolution processes. Currently, in situ tests and laboratory tests are the main approaches to obtain hydraulic conductivity. However, both of these methods have obvious drawbacks (e.g., size and location limitations); thus, the test results cannot be directly adopted for solving many geotechnical engineering problems that are of great inhomogeneity. To overcome these drawbacks, a back-analysis approach is proposed in this paper to obtain the saturated hydraulic conductivity based on groundwater observation. This approach includes three main steps: (1) develop a saturated–unsaturated seepage model based on field investigation; (2) use the Generalized Regression Neural Networks method to establish a non-linear mapping relation between saturated hydraulic conductivities and groundwater levels; and (3) employ the Genetic Algorithm method to search for the optimum solution for hydraulic conductivity under which the calculated groundwater levels in the seepage model fit best with the observational groundwater levels. Furthermore, this approach is applied to back analyze the hydraulic conductivities of the riverside slump I# in the Huangtupo landslide, which is volumetrically the largest reservoir landslide in the Three Gorges Reservoir area in China. The observational groundwater levels that are used to back analyze the hydraulic conductivities are the result of a response of the whole landslide to the reservoir water level change. Consequently, this approach overcomes the aforementioned limitations in the tests, and the results provide more reliable references for studying reservoir landslides.