Abstract: This paper presents advanced techniques for the detection of bearing failures based on vibration signals. Capability of detection and diagnosis of some effective techniques are discussed and compared on the basis experimental results. In particular, we analyzed the ID3 data mining technique, detection using statistical pattern recognition and application of hidden Markov model (HMM). Comparing the experimental results showed that the highest accuracy of detection techniques has been realized using HMM, and for determining the type of bearing damage easier to use a statistical approach to pattern recognition. It is shown that the presence of faults can be detected on-line monitoring of the probability of trained HMM for the correct bearing, which is determined by feature extraction from vibration signals. In addition, the application of HMM can be extended to consider problems of prognosis and prediction of bearing condition, which gives as an output time to failure.