Measuring in vivo dynamics can yield valuable information about the cardiovascular or musculoskeletal system. MRI shows severe limitations when dealing with motion at high spatial and temporal resolutions. We propose spectro-dynamic MRI for the characterization of dynamics directly from k-space data. The sampling pattern is adjusted to achieve a temporal resolution of a few milliseconds. A measurement model, relating the measured data to the displacements, is combined with a dynamical model, introducing prior knowledge about the dynamics. Preliminary results show that spectro-dynamic MRI allows estimation of motion fields and dynamical system parameters from heavily undersampled data on a millisecond timescale.