As personal multimedia devices become mainstream consumer products, an energy-aware on-demand video streaming system for portable devices is critical. A critical component of an energy-aware video streaming system is an accurate energy prediction scheme for video decoding. In this paper, we propose the first content-aware energy predictor for video playback. We correlate video attributes with decoding energy through linear regression. The average prediction error achieved by proposed content-aware energy prediction scheme is only 1.88%.