In fiber gas sensor systems, in order to improve the detection sensitivity, it is vitally necessary to make intensive efforts to restrain various noises of the system. Designing low-noise preamplifier and light driving circuit can effectively reduce the system noise, but cannot reduce the noise caused by the changes of optical path. This kind of noise is called background noise, which influences the measuring sensitivity of fiber gas sensor system. Therefore, how to filter the background noise affects the improvement of measurement accuracy. This paper puts forward a method based on least logarithm square method which can easily reduce the background noise of optical gas sensor system. By using reference grating and Fabry-Perot etalon, it can reconstruct the output spectra. A modified least square method is used to fit the envelope of gas absorption lines by using the data of absorption peaks on both sides. Dividing the gas absorption lines by the envelope line of absorption lines, the normalization of output is realized and thus the drift of the background noise curve due to the change of experimental environment can be reduced. This method can avoid data collection again when the background noise is drifting, and brings convenience to measurement.