In order to effectively reduce noise in vibration signals, the principle of total variation denoising method is introduced from image processing. The effect of the parameter λ in the noise reduction algorithm is analyzed, and the λ selection method is proposed based on kurtosis and the cross-correlation coefficients of signals and denoised signals. The balance between the fidelity constraint and denoising is established. A simulation signal and a bearing vibration signal are used for verification. The results show that the total variation-based noise cancellation method proposed here is feasible and valid.