We present a system for rhythmic analysis of human motion inreal-time. Using a combination of both spectral (Fourier) andspatial analysis of onsets, we are able to extract repeating rhythmic patterns from data collected using accelerometers. These extracted rhythmic patterns show the relative magnitudes of accentuated movements and their spacing in time. Inspired by previouswork in automatic beat detection of audio recordings, we designedour algorithms to be robust to changes in timing using multipleanalysis techniques and methods for sensor fusion, filtering andclustering. We tested our system using a limited set of movements,as well as dance movements collected from a professional, bothwith promising results.