A signal denoising method based on short-time Fourier transform (STFT) time-frequency spectrum coefficient shrinkage is presented in light of the denoising of rotating machinery fault vibration signals. The STFT is adopted to transform the target signal into a time-frequency domain, and the spectrum′s complex coefficients are shrunk accordng to their modulus magnitude. A step iterative algorithm is proposed to estimate the optimal threshold at the interval between 0 and the maximum coefficient modulus. First, both the traditional hard and soft threshold functions are used for coefficient shrinkage. The optimal threshold estimation can then be obtained according to the modified risk function. Finally, the inverse STFT is applied to the spectrum coefficients after they are shrunk with the optimal threshold, and the obtained time-domain denoised signals are reconstructed. The results of the emulational signal and experimental data have demonstrated that the STFT time-frequency spectrum coefficient shrinkage method with either the hard or soft threshold function can work well in vibration signal denoising with the estimated optimal threshold.