With the development of the economic, the problem of the global greenhouse effect has been attached more importance than before. In order to predict the emission of the carbon dioxide in the coming years more precisely and take corresponding energy conservation and emission reduction measures in time. In this paper, the genetic algorithm (GA) is used to optimize the initial connection weights and thresholds of the tradition Back Propagation Neural Network (BPNN) named GA-BPNN model, which can overcome BPNN’s shortcoming of converging to the local minimum obviously. The data of China during the period 1985-2015 including carbon emission and influence factors are selected to perform the carbon dioxide emission prediction with the established model. The results indicates that the GA-BPNN model established in this paper has a higher accuracy, which is more applicable to the current prediction of carbon emissions.