Fault diagnosis in bearings has been the subject of intensive research as bearings are critical component in many-rotating machinery applications and an unexpected failure of the bearing may cause substantial economic losses. Vibration analy-sis has been used as a predictive maintenance procedure and as a support for machinery maintenance decisions. In this study, we used Multiclass Support Vector Machines (MSVM) for classifica-tion task. SVM is a powerful classification tool that is becoming increasingly popular in various machine-learning applications. In data pre-processing step advance signal processing methods such as wavelet transform is used to extract the features which reveals the characteristics of the actual conditions of the bearing. To reduce the dimensionality of the feature SVM-Recursive Feature Elimination (SVM-RFE),Wrapper subset method, ReliefF meth-od and Principle component analysis (PCA). Moreover, investiga-tion is done performance of MSVM with mentioned different feature selection strategies as well as MSVM alone on basis of classification accuracy. To validate the proposed methodology, four kinds of running states are simulated with artificially creat-ed faults in bearing that is accommodated in the test rig. By com-paring classification accuracy along with computation timing the best scheme is selected for diagnostic prediction. © February 2015 IJENS.