The main solid waste of metal mine are tailings. The mixture of tailings and cement can be made into filling slurry to fill the mined out space, which is a typical method of solid waste treatment. The prediction for mix proportion of filling slurry is a complex nonlinear process, and the strength of filling body formed by filling slurry is very significant to cemented tailings backfilling. Based on the introduction of RBF neural network, learning samples of network model were obtained through laboratory experiments, and the model accuracy was tested via predictive samples. With in situ engineering case, the predictive results of model and practical measurements are compared and verified. Testified by practice, using RBF neural network to predict the mix proportion of filling slurry is feasible.