To effectively identify the overlapped peaks and diminish the interference of random noise in laser-induced breakdown spectroscopy (LIBS), an automatic peak detection method using Gold deconvolution algorithm combined with smoothing is proposed. For the LIBS of standard samples NIST 1256b and NIST 1762, smoothing is performed first, followed by deconvolution using Gaussian system function, and finally, five-point method is used to detect the peak automatically from the deconvolution results. Thus, the effect of smoothing combined with system functions with different full width at half maximum (FWHM) on the ability to identify overlapped peaks and reduce the interference of random noise is studied experimentally. The results show that when the FWHM of 0.140 nm (slightly larger than that of weak peaks) is used, combined with smoothing, the resolution of Gold deconvolution algorithm to identify overlapped peaks can be up to 0.039 nm, and the wavelength, intensity of the peaks, as well as curve fitting, are well matched with NIST database. The preliminary results indicate that the automatic peak detection based on Gold deconvolution algorithm can be applied to LIBS data processing.