A new discrimination algorithm of Raman spectral peaks is proposed. With Voigt function as the fitting function, the spectral peaks obtained after the ridge peak detection based on the wavelet transformation are fitted. The similarity of spectral peaks before and after fitting is taken as the basis for discrimination of spectral peaks. The Raman signal with amplitude comparable to that of noise is effectively discriminated. Compared with those of the existing two discrimination methods of spectral peaks, the true positive rate of the proposed method improves by 60% when the false data rate is 5% and the false data rate of the proposed method decreases by 10% when the true positive rate is higher than 90%.