This paper is focused on the peak-detection algorithms in the demodulation for the fiber Bragg grating (FBG) sensor system. 6 peak-detection algorithms have been analyzed and compared, such as the Monte-Carlo algorithm, the direct peak-located algorithm and the quadratic polynomial numerical derivative algorithm, the polynomial fitting, the polynomial-Gaussian fitting and the Gaussian nonlinear curve fitting. The theoretical and practical errors and the relative effect factors of errors were introduced, analyzed and evaluated by the combination of the simulations and the experiments. It is demonstrated that the relationship between the signal noise ratio (SNR) at the input of the algorithms and the error is linear. When the SNR is constant, the error in using Gaussian nonlinear curve fitting is the lowest. The error can be only 0.44 pm when the SNR is 40 dB in the FBG sensor experiment. Consequently SNR is the major factor which dominates the errors of the peak-detection algorithms in the demodulation and the Gaussian nonlinear curve fitting is considered as the best peak-detection algorithm.