The stress-strain behavior of hot-compression Mg-Li alloy was investigated by using a physical simulator Gleeble-3500 system. And the constitutive equation was set up by regression analysis and BP neural networks. Results show that the dynamic recrystallization occured during the hot-compression process. The grain size of the alloy increased and the stress decreased with increasing temperature. Regression analysis indicates that the flow stress can be expressed by hyperbolic sine model and the arithmetic average of errors is 14.13%. Training the flow stress prediction model with MatLab by an improved BP,the maximum arithmetic average of errors is 4.27%. The predicted stress-strain curves are in good agreement with the experimental results.