Abstract: A backward analysis of physical and mechanical parameters of surrounding rock was newly developed, based on differences method, orthogonal design and artificial neural networks, where the different parameters were chosen according to orthogonal design. The related analytical samples for neural networks were given by FLAC calculation results. With the samples trained, the structure and learning parameters of networks were optimized. The physical and mechanical parameters of surrounding rock of Hanjialing tunnel were analyzed backwards by trained results and the on-the-spot surveyed data. The results satisfied the demand of precision.