A novel approach for the compression of mechanical vibration signals is presented in this paper. The method relies on a simple and flexible decomposition into a large number of subbands, implemented by an orthogonal transform. Compression is achieved by a uniform adaptive quantization of each subband. The method is tested on a large number of real vibration signals issued by plane engines. High compression ratios can be achieved, while keeping a good quality of the reconstructed signal. It is also shown that compression has little impact on the detection of some commonly encountered defects of the plane engine.