Cost prediction can assist managers in cost management. There is no standard prediction method at present. In this paper, the MLP (multi-layer perception) is used to establish the cost prediction model, and the MLP neural network in SPSS is used for analysis. The accuracy of MLP prediction value and BP (Back Propagation) neural network prediction value is compared. It is found that for R2, the accuracy of MLP is improved by 11%. The results show that MLP is more suitable for the prediction of construction cost. It improves the accuracy of project cost prediction and provides a new idea for cost prediction.