The analysis of the sampled fiber Bragg grating reflection spectrum usually obtained with so much noise. Levenberg-Marquardt (LM) algorithm is used to optimize the Gaussian curve fitting coefficient and the optimized Gaussian-LM algorithm is used for fiber Bragg grating spectrum peak detection. The comparison of the peak search results proves that the Gaussian-LM algorithm is important for optimizing the Gaussian curve coefficient. Research shows that Gaussian-LM peak detection algorithm can effectively reduce the peak error. It also improves stability of central wavelength detection and accuracy in fiber Bragg grating sensor system based on wavelength demodulation system.