Abstract: In order to improve the reliability of numerical simulation results, a new backward analysis method for mechanical parameters of slope rocks was developed based on orthogonal design, difference method and artificial neural network. According to orthogonal design, the value levels of the mechanical parameters were chosen, and simulation schemes were arranged; the related analytical samples for neural network were given by FLAC^2D calculations; RBF neural network was trained; the physical and mechanical parameters of an open pit slope were analyzed backwards by well-trained RBF neural network and surveyed data about spot displacements. The error between the backward analysis results and the theoretical ones is much little and meets the demand of precision, which indicates that this backward analysis method is feasible and accurate.