A demodulation algorithm is vital for a fiber Bragg grating (FBG) sensing system. In this paper, a novel demodulation algorithm based on the variable-step-size method and crosscorrelation algorithm is proposed to demodulate the wavelength of an FBG. By changing the step size of each calculation, the amount of calculation can be greatly reduced, and pm-level resolution can be achieved over the whole bandwidth. A simulation shows that the demodulation results are not affected by noise, and the algorithm can accurately demodulate the FBG wavelength in a large range. The stability of the algorithm is also tested using three FBG sensors, and the results show that the difference among 25 measurements is only 4 pm. Moreover, the results of temperature calibration experiments show that the sensitivity of the sensor is 41.9 pm/℃, which is similar to the value of 45 pm/℃ provided by the manufacturer. The R 2 value of the fitting curve is 0.9994, which means that the demodulation algorithm can be effectively used in sensing measurement.