Automatic peak seeking is an indispensable link for in situ and real-time spectral detection and analysis, and has important significance for application of spectral technology to such fields as long-term marine monitoring and oil mud logging. Based on some typical LIBS/Raman spectrum data obtained from lab, three kinds of symmetric zero-area transformation functions respectively constructed from Gaussian, Lorentz and Voigt function were compared, and the results show that there exists an optimal symmetrical zero-area transformation function for peak seeking, but all the transformation functions obtain the same peak position and peak width under their optimal parameters. The proposed method is free from spectrum background and baseline trend influence, adaptive to the wide range of SNR, close to or even better than artificial recognition for weak peak, and could be used in future automatic in-situ analysis of LIBS and Raman.