Abstract: Mine construction has the characteristics of huge investment, long cycle, large uncertainty, which determine the speed and accuracy of the investment estimate for investment decisions is essential. Established investment estimation model based on 10 estimates subsystems, for example bottom drift, selecting the section size, supporting way, rock bolt consumption and other technical and economic indicators as engineering features, making quantification and normalization, provided neural network function with Matlab to build BPNN (Back Propagation Neural Network)model for predicting the roadway project investment, the model included three layers, seven input indicators, one output indicators. The results show that, as long as the selecting of project characteristics and BPNN model parameter settings were appropriate and accurate, the neural network method can quickly achieve the target, the prediction accuracy could reach ± 10% or less, meeting requirements of estimates predict rapidity and accuracy.