Wireless sensors with accelerometers are widely used in various studies on human body movements. The most challenging problem in a small body-attachable sensing unit is how to maximize the battery lifetime. Previously, the preferred approach was to reduce the number of transmissions through data compression. Compressed Sensing(C-S) is an emerging alternative approach that aggressively reduces the samples yet permits the reconstruct of the original analog signal. C-S has the great potential to be extremely effective due to the universality and lower complexity of sensor implementation. In this paper, we investigate the nature of various human body movements. We examine the performance of the C-S framework in terms of the energy savings in a real testbed. Our experimental results show that the C-S framework can save up to 40% of energy in the sensing unit, compared with the traditional data compression scheme.