When the compressed sensing theory is used to solve the noise reduction problem of vibration signal,the orthogonal matching pursuit algorithm is most used in the signal reconstruction stage.If the iteration stop threshold of the algorithm is not properly selected,the cycle of iteration error will lead to the increase of reconstruction error and the decline of noise reduction performance.Therefore,a noise reduction method was proposed to improve the stopping threshold criterion.The discrete cosine transform was used to obtain the complete dictionary matrix,and the sparse coefficient vector was obtained by sparse representation of vibration signal.Then the normal distribution was used to test whether the sparsity coefficient obeyed the normal distribution,and the stopping threshold was calculated by using the 3σ criterion.Finally,the screening condition and iteration stopping threshold were added after the step of solving the least squares solution.The noise reduction analysis results of simulated defined signal and measured vibration signal show that the improved method can improve the signal to noise ratio (SNR) of strong noise background signal from -1.025 7 dB to 2.054 9 dB.The proposed method is superior to SP,OMP and SWOMP in terms of noise reduction performance indicators such as SNR and root mean square error.