An intelligent diagnosis for the faults of motor bearing is the emergence of fluctuations in the field of fault diagnosis, which represents a sequential method on the basis of vibration signals and soft margin support vector machines (SVMs). Vibration signals are extracted the features of bearing faults to purify symptom parameters (SPs) in frequency domain, and sensitive SPs are selected by discrimination index (DI). Multitude soft margin SVMs are used sequentially to detect fault and identify fault type. The presented method has been applied to diagnose bearing faults in real plant, the experimental results shows that the proposed method is efficient.