Valuable achievements on differential optical absorption spectroscopy (DOAS) for monitoring atmospheric pollutants gas have been made in the past decades. Based on the idea of setting the threshold according to the maximum value, symbolized as OD′m, of differential optical density, the algorithm of traditional DOAS was combined with the DOAS algorithm based on the kalman filtering to improve the detection limit without losing measurement accuracy in the present article. Two algorithms have different inversion accuracy at the same ratio of signal to noise and the problem of inversion accuracy was well resolved by combining two algorithms at short light path length. Theoretical and experimental research on the concentration measurement of SO2 in the flue gases was carried out at the normal temperature and atmospheric pressure. The research results show that with the OD′m less than 0.048 1, the measurement precision is very high for SO2 with the improved DOAS algorithm. The measurement lower limit of SO2 is less than 28.6 mg·m-3 and the zero drift of the system is less than 2.9 mg·m-3. If the OD′m is between 0.048 1 and 0.927 2, the measurement precision is high with the traditional DOAS algorithm. However, if the OD′m is more than 0.922, the errors of measurement results for both two DOAS algorithms are very large and the linearity correction must be performed.