Abstract: The signal collected by using the spectrometer is inevitably affected by different noise sources. In order to increase the analytical accuracy of spectrum signal, an improved denoising method with threshold was presented by analyzing the principle of applying the wavelet to signal denoising and the defects existed in classical denoising methods by using soft and hard thresholds. The improved method can overcome both the defects，that the hard threshold method generates discontinuities and the soft threshold method generates a constant variation, and can retain the useful signals as much as possible. In experiment, the SymletsA wavelet was selected as the wavelet function with 4 series, through combining with the perlevel threshold determined by the BirgeMassart tactics pattern,the denoising process was conducted for the fluorescence spectrum signal of CdSe quantum dots . Simulation results demonstrate that the rebuilt signal by the improved denoising method with threshold can has a increased signaltonoise ratio(SNR=47.550 2) and a energy accounting(PER=0.973 3), as well as an improved mean squared error(MSE=149.421 3) compared with the classical denoising methods by using soft and hard thresholds.