The forecast of the compressive strength of ultra-high-performance concrete (UHPC) is important for engineering design, and it is of great significance to construction safety. Currently, neural networks-based methods have made great progress in forecasting UHPC compressive strength, and GA-BP neural networks and recurrent neural network (RNN) are commonly used forecasting models. However, for UHPC compressive strength prediction system, the comparative study between the forecasting time and system stability of the two methods is still lacking. In the paper, a UHPC compressive strength prediction system was built based on GA-BP neural network and RNN and a comparison was conducted. The prediction performance was evaluated by RMSE and the forecasting time and the system stability were compared. The UHPC compressive strength prediction system based on the GA-BP neural network achieved better results on the RMSE index with lower prediction error.