When motor bearing faults occur, the weak fault features in the stator current signal are drowned in the strong noise background of fundamental waves and harmonics. Because of the low signal-to-noise ratio, current-based bearing fault detection has always been a challenge. To this end, this paper proposes a detection method based on time shift denoising and a cyclic bispectrum. First, the stator current analytical expression is constructed according to the cyclic bispectrum function, and the characteristic solution of the current analytical expression is extracted. Then, the unrelated noise component in the stator current signal is decomposed, and the delayin the cyclic bispectrum function is quantitatively analyzed, and the optimal time shift is determined. Finally, the stator current is collected by the experimental platform, and slice spectrum analysis is performed at the position where the current solution is used to obtain the characteristic solution. The experimental results show that the proposed method can effectively improve the signal-to-noise ratio and extract non-stationary weak fault features in the current to realize the detection of motor bearing faults. This work is supported by National Natural Science Foundation of China (No. 51279020).