It is important to decrease the statistical error for improving the measurement precision of Raman lidar for the large statistical error of the Raman backscattering signals. An adaptive filter signal processing technique is applied to the Raman lidar for atmospheric CO2 measurement by subsection,and the result of the statistical error of the atmospheric CO2 mixing ratio changes little with detecting range in each subsection. After signal processing by the technique,we obtained the high precision of Raman lidar measurement of atmospheric CO2 concentration at night in Hefei. The maximum statistical error of atmospheric CO2 concentration was 2.5% from 1.5 to 5 km,5% form 5 to 8 km,and 10% from 8 to 10 km. The technique also can be used to determine the number of laser shots necessary for a desired statistical accuracy and spatial resolution.