Regression based methods are widely used for 3D and 2D human pose estimation, but the performance is not satisfactory. One problem is that the structural information of the pose is not well exploited in the existing methods. In this work, we propose a structure-aware regression approach. It adopts a reparameterized pose representation using bones instead of joints. It exploits the joint connection structure and proposes a compositional loss function that encodes the long range interactions in the pose. It is simple, effective, and general. Comprehensive evaluation validates the effectiveness of our approach. It significantly advances the state-of-the-art on Human3.6M and achieves state-of-the-art results on MPII, in a unified framework for 3D and 2D pose regression.