This document presents a new approach for dense non-rigid structure from motion (NRSfM) using monocular video. Previous NRSfM methods provided sparse 3D reconstructions, while this new approach enables dense 3D reconstruction for every frame. It formulates NRSfM as a variational energy minimization problem that estimates projection matrices and 3D shapes by minimizing reprojection error with spatial smoothness and low-rank priors. The energy is optimized iteratively to estimate rotations and refine shapes. Results on real sequences demonstrate improved accuracy over previous sparse NRSfM methods.