The development of economy, the progress of the society and the modern civilization is inseparable from the energy. Along with rapid development of economy and society, the energy demand grows continuously. Therefore energy demand forecasting has important theoretic and realistic significance. The BP neural network is usually applied to energy demand forecasting. But traditional BP neural network easily gets into part minimum, which leads to non convergence of algorithm and fail training. The master slave neural network (MSNN) is consisted with two Hopfield networks as master network and a BP network as slave network because of its good dynamic evolution performance. It can solve the problem well. Compared with BP neural netwok, MSNN has smaller system error and quicker asymptotic convergence rate. This paper proposes the energy forecasting new model and it is applied to predict energy demand in the last five years. The result shows that MSNN has a more rapid convergence rate. Besides it has smaller network system errors. It predicts ultimately energy demand well. Therefore, the MSNN can improve effect of energy demand forecasting better.