Abstract: The energy demand system of Beijing has a number of features including nonlinearity, limited historical data and numerous drivers, while the Support Vector Machine(SVM) model owns unique advantage in small samples, nonlinear and high-dimensional pattern recognition. Therefore, the SVM model is employed to fit the related historical data about energy demand in Beijing from 1978-2010 and then project the energy demand during 2012-2020. The results indicate that the projection power of SVM model evidently outweighs that of other traditional and commonly-used models, which can effectively consider the complex features in energy demand system of Beijing. Additionally, the projection results also suggest that Beijing’s energy demand may increase year by year during 2012-2020, with the average annual growth rate 2.75%; and its growth rate may appear relatively slower in the 13th Five-Year Plan period than that in the 12th Five Year Plan period.