Based on the hot compression test data of as-cast AZ80 magnesium alloy under the conditions of deformation temperature of 250~400 °C and strain rate of 0.001~1 s^-1, a physical-based constitutive model based on the stress dislocation correlation and dynamic recrystallization dynamics and an artificial neural network (ANN) model based on feedforward backpropagation algorithm were established to predict the thermal deformation behavior of AZ80 magnesium alloy. Three statistical indicators, correlation coefficient (R), mean absolute relative error (AARE), and relative error (RE), were used to verify the prediction accuracy of these two models. The results show that both the models can accurately predict the thermal deformation behavior of AZ80 magnesium alloy. The stress value predicted by ANN model shows better agreement with the experimental data, and the value of R and AARE of ANN model is 0.9991 and 2.02%, respectively. While the R and AARE predicted by the physical-based constitutive model are 0.9936 and 4.52%, respectively. The better predictive ability of ANN model is attributed to its ability to deal with complex nonlinear relationships, while the predictive ability of the physical-based constitutive model is attributed to the fact that the model has certain physical meaning. The thermodynamic mechanism of work hardening (WH), dynamic recovery (DRV), and dynamic recrystallization (DRX) during thermal deformation are fully considered in the model parameters. Finally, the advantages and disadvantages of these two models are compared and discussed.