Abstract: Based on the testing results of 5 joints at a steel reinforced high-strength concrete column/reinforced high-strength concrete girder frame, some P-▵ curves have been generated under the action of low-frequency cyclic load. Based on the principle of BP neural network, a model is developed by way of setting up the input, implicit and output layers with input and output cells and number of nodes or joints of implicit level all determined. Then, the network is trained up according to some given data to enable it to serve the functions of analysis and judgement so as to forecast the P-Δ curves generated from the joints of a steel reinforced high-strength concrete frame. The results show the feasibility of the method proposed.