Abstract: In order to shorten the cycle of filling experiment and find the optimal raw material ratio as soon as possible, the results of orthogonal experiment and single factor experiment were used as the training samples of BP neural network, and the percentage of ash in the different raw materials and the ash-sand ratio were used as input parameters, and the slurry expansion, 7 d and 28 d backfill strength of the filler were used as output results, the BP neural network model was establised to predict the strength of steel slag tailings-slag based filler. The established model was used to optimize the preparation of filling body under the condition of 16.0% steel tailings ratio. The results show that at 14 neurons in the hidden layer, the maximum relative errors of the flow and backfill strength in 7, 28 d are 0.36%,1.46%,2.23% respectively, and the average relative errors of backfill strength of fillers out of samples in 7, 28 d are 2.86%, 1.36% respectively. The errors inside and outside the sample are small, the prediction accuracy of the model is high and the generalization ability is strong. For the filler prepared by modle optimization, the predicted value of 28 d backfill strength is 2.16 MPa, and the measured value is 2.13 MPa, and the difference is small. The BP neural network can be used to predict the backfill strength, which has guiding significance for optimizing the ratio of filling raw materials and reducing the workload of filling experiment.