Video frame interpolation (VFI) aims to generate predictive frames by warping learnable motions from the bidirectional historical references. Most existing works utilize spatio-temporal semantic information extractor to realize motion estimation and interpolation modeling, not enough considering with the real mechanistic rationality of generated middle motions. In this paper, we reformulate VFI as a multi-variable non-linear (MNL) regression problem, and a Joint Non-linear Motion Regression (JNMR) strategy is proposed to model complicated motions of inter-frame. To establish the MNL regression, ConvLSTM is adopted to construct the distribution of complete motions in temporal dimension. The motion correlations between the target frame and multiple reference frames can be regressed by the modeled distribution. Moreover, the feature learning network is designed to optimize for the MNL regression modeling. A coarse-to-fine synthesis enhancement module is further conducted to learn visual dynamics at different resolutions through repetitive regression and interpolation. Highly competitive experimental results on frame interpolation show that the effectiveness and significant improvement compared with state-of-the-art performance, and the robustness of complicated motion estimation is improved by the MNL motion regression.