Abstract: Evaluating the mechanical performance of polymer modified mortar (PMM) is essential to ensure safe usage. To efficiently and accurately obtain PMM with excellent mechanical properties, a backpropagation neural network (BPNN) prediction model with a topology structure of 6-14-2 was designed and optimized using a genetic algorithm (GA). The input layer of the GA-BPNN model comprised cement, cellulose ether, dispersible polymer powder, antifoam, limestone powder, and fly ash content, with the output layer encompassing compressive strength and bond strength. The dataset consisted of 520 data points, of which 60% was used to build the model and 40% to validate the model. Experimental flexural strength, compressive strength, and bond strength were taken as the evaluation indicators for PMM's mechanical properties. The relationship between raw materials and PMM mechanical properties was determined through correlation matrix analysis and principal component analysis, along with comparative analysis of mechanical performance indicators. The results indicate that at 7 d and 28 d, the development of PMM mechanical properties is positively correlated with dispersible polymer powder and antifoam. At 7 d, limestone powder and fly ash are negatively correlated with compressive and flexural strength, while cellulose ether is positively correlated with bond strength. At 28 d, cement is negatively correlated with compressive, bond, and flexural strength, and positively correlated with limestone powder and fly ash. The GA optimization algorithm can significantly enhance the prediction accuracy of the BPNN model. The predictive performance indicators of GA-BPNN for compressive strength and bond strength are R² = 0.918、R_MAE = 17.507、R_MAPE = 0.299、R_RMSE = 7.849; R² = 0.922、R_MAE = 17.101、R_MAPE = 0.282、R_RMSE = 8.077. Thus, GA-BPNN can provide accurate predictions for the mechanical performance of PMM and guide its mix design, which is of great significance for engineering practice.