Three peak-detection algorithms (cubic spline interpolation differential method, Gauss polynomial, Gaussian fitting algorithm) for fiber Bragg grating reflection spectrum were analyzed and compared. It is demonstrated that Gaussian fitting algorithm produces the lowest error under the same sampling conditions; when the uniform sampling number is 250, the errors of three peak-detection algorithms are respectively 3.4 pm, 13.0 pm and 2.6 pm. The idea of non-uniform data sampling method to search peak was proposed. Sample the actual FBG reflection spectrum with uniform and non-uniform sampling method, and search peak using the three peak-detection algorithms respectively. It can be concluded that the error of a certain algorithm by non-uniform sampling is lower than by uniform sampling method; when the sampling number is 250, FBG peak error of Gaussian fitting algorithm under non-uniform sampling method is reduced by 38.46% than uniform sampling method.