IEEE Bearing failure often occurs in doubly-fed induction generator (DFIG)-based wind turbines which are usually subject to electrical corrosion effects. Fault diagnosis method based on electrical signals has been paid much attention as the method is non-invasive and cost-effective. This paper describes the use of the modulation signal bispectrum (MSB) detector for diagnosing bearing faults in DFIGs of wind turbines. The major theoretical principles involved with the MSB method are presented and it is shown how the amplitude and phase relationships of the stator current signals caused by torque oscillations can be effectively revealed. Since the MSB result is obtained by averaging results from each record, overlapped segmentation is proposed to improve computational accuracy with limited data. On site experimental results obtained from 1.5 MW wind turbines corroborate that these faults can be detected, in the current MSB, by the identification of a spectral component at the fundamental frequency and the characteristic frequency. Compared with the other data processing methods based on second-order cumulants, the MSB detector can avoid misdiagnosis by containing phase information of stator current. Owing to relatively high accuracy, the proposed current based MSB method can identify incipient bearing corrosion failure in DFIG-based wind turbines without additional sensors, which also has great potential in other industrial applications.