Correlation coding technology can effectively improve the signal-to- noise ratio of a fiber Raman distributed temperature sensor system.However,the traditional decoding algorithm in this t echnique needs a lot of calculation and memory space.To solve the two problems,a method of overlap saving is proposed to cut Raman scattering signals into segments.Furthermore,the traditional discrete line ar correlation decoding algorithm is replaced by the circular correlation decodi ng algorithm which is based on the fast Fourier transform (FFT) in each segment.Finally,the results of each segment are combined together to complete the decoding of the system.In the simulation of a 20km fiber R aman distributed temperature sensor system which is based on 512bit Golay pulse coding sequence,the time of the circular correlation decoding process can be reduced by more than eight times from 0.228s to 0.025s compared with the traditional disc rete linear correlation decoding process.In the mean time,the memory space of the FFT module can be reduced by more than 19time s from 1280000bit to 65536bit compared with the traditional discrete linear correlation decoding process.The above results indicate that the decoding method is efficient and can be performed in real time.