Features in time and frequency domain were extracted first. A compensation method based on distance was used to choose features sensitive to bearing fault. Then full features and sensitive features vectors were built. The results of Hidden Markov Model based on those two features were different. Then the method of integrated HMM for bearing fault diagnosis was proposed. Based on independent HMM classifiers trained by those two different features, average rules and the maximum likelihood probability method were used to combine the HMM classifiers. The experimental results show that this method has a higher recognition rate compared with the other two independent classifiers based on different feature vectors.