To minimize demodulation error and system instability due to the Guassian white noise and power frequency noise, we presented a fiber Bragg grating (FBG) peak-detection algorithm based on difference of Gaussian (DoG). This algorithm enabled the detection of the peak wavelength of FBG reflection spectrum with strong noise. The mean error and standard deviation are 4 pm and 4.2 pm, respectively, which are much lower than those of other traditional algorithms. Experimental results show that the DoG algorithm can provide good stability and low demodulation error. In addition, it can substantially resist the noise in the FBG reflection spectrum, thereby improving the accuracy of the FBG demodulation system.