The purpose of this thesis was to implement and test a wavelet based feature extraction method for classification of bearing signals. A specific method, developed by Saito, was thoroughly investigated and implemented in Matlab. This algorithm has been tested on accelerometer signals originating from faulty and functioning bearings. Signals from both laboratory and industrial environments has been analyzed. It has been shown that this method can be used to find faulty bearings if the algorithm has been trained with signals from similar, both functioning and faulty, bearings. Provided these circumstances, it seems as if this method could be more effective, for some cases, than existing FFT methods.