Abstract: A new method based on Radial Basis Function Neural Network (RBFNN) is proposed to improve the accuracy of existing calculation method for key points (ultimate stress and strain) in stress-strain model for FRP-confined concrete. In RBFNN model, concrete axial strength, tensile strength of FRP, FRP volumetric ratio, corner radius-to-section width ratio and aspect ratio were considered as input factors, and the compressive strength ratio and ultimate strain ratio were adopted as output factors. Trained by existing experimental data, RBFNN with highly non-linear reflection relationship was founded and proved to be more effective and accurate in calculating the key points in stress-strain model. Combining RBFNN method with the existing stress-strain model, an improved calculation method is put forward to predict stress-strain curves, the calculated results show reasonable agreement with other test results.