Abstract: High-temperature compression tests of GH4065 alloy were carried out under conditions of deformation temperature of 1 150–1 350 K and strain rate of 0.001–1 s⁻¹. The true stress-true strain curves were obtained. By the back propagation (BP) algorithm, the neural network model for high-temperature deformation of the alloy was constructed, and the topological structure of the model was determined by self-optimizion method of neural network structure. The accuracy of the BP neural network model in predicting the high-temperature deformation behavior of the alloy was verified, and was compared with that of Arrhenius variable parameter constitutive model considering true strain effect. The continuous mapping relationships among the true strain, deformation temperature, strain rate and true stress were established by the BP neural network model. The results show that the BP neural network model had a 3-10-11-1 structure. The average relative errors between the true stress predicted by the training and validation data and the test results were 3.54% and 2.77%, respectively, and the correlation coefficients were 0.997 4 and 0.994 8, respectively, indicating that the BP neural network model could accurately predict the high temperature deformation behavior of the alloy. The prediction accuracy for the true stress of GH4065 alloy during high temperature deformation of BP neural network model was higher than that of the Arrhenius variable parameter constitutive model ,whose average relative error was 9.54% and correlation coefficient was 0.979 1.