Plasma emission spectroscopy is one of the important diagnostic methods to study the physics of plasma within tokamak. There are complex atomic and molecular physical processes in the boundary plasma, and some weak particle spectrum signals are mixed witha lot of noise. Therefore, whether the noise can be effectively removed and the signal quality can be improved is of great significance for subsequent analysis and understanding of related physical processes in experiments. Taking the simulated signals and the tungsten atomic spectrum data in tokamak experiment as research subjects, the denoising effect of wavelet threshold denoising method is studied in this work by using the signal-to-noise ratio(SNR) and root mean square error(RMSE) as the judgement basis of filtering effect. The comparative analysis of simulation experiments shows that when sym8 wavelet base, 4-layer wavelet decomposition, heuristic threshold calculation and progressive semi-soft threshold function are selected for wavelet denoising, the maximum signal-to-noise ratio of 19.2166 and the minimum root mean square error of 0.0290 can be obtained. Furthermore, the as-received approach and the best matching parameters are applied to the signal processing of measured divertor tungsten atomic spectrum, and a good denoising effect is also obtained. So, it is validated that the wavelet threshold denoising can effectively eliminate the noises in the tungsten atomic spectrum while avoiding signal distortion, that is, the method can significantly improve the signal quality.