The wavelet threshold denoising method is an effective noise suppression approach for noisy laser-induced breakdown spectroscopy signal. In this paper, firstly, the noise sources of LIBS system are summarized. Secondly, wavelet multi-resolution analysis and wavelet threshold denoising method are introduced briefly. As one of the major factors influencing the denoising results in the process of wavelet threshold denoising, the optimal decomposition level selection is studied. Based on the entropy analysis of noisy LIBS signal and noise, a method of choosing optimal decomposition level is presented. Thirdly, the performance of the proposed method is verified by analyzing some synthetic signals. Not only the denoising results of the synthetic signals are analyzed, but also the ultimate denoising capacity of the wavelet threshold denoising method with the optimal decomposition level is explored. Finally, the experimental data analysis implies that the fluctuation of the noisy LIBS signals can be decreased and the weak LIBS signals can be recovered. The optimal decomposition level is able to improve the performance of the denoising results obtained by wavelet threshold denoising with non-optimal wavelet functions. The signal to noise ratios of the elements are improved and the limit of detection values are reduced by more than 50% by using the proposed method.