Differential optical absorption spectr oscopy (DOAS) has been widely applied in the flue gas concentration determination based on Lambert-Bee r′s law. However,under high concentration pollutants,the bias from Lambert-Beer′s law increases due to the pollutant molecule interactions,and c auses inevitable errors only with least squares regression.Therefore,a non-linear model fo r flue gas concentration determination based on standard deviation of differential optical density is put forward in this paper,proved by both single-component and mixture experiments with standard SO_2and NO,and compared with traditional least-squares (LS) modeling method.T he results of non-linear modeling method indicate that in the single-component experiment,t he correlation coefficients (R²) of non-linear models are 0.9996for SO_2and 0.9995for NO,and the maximum errors greatly reduce to 2.6% and 2.8% compared with －16.1% and －19.9% using traditional LS modeling method; in the mixture experiment,the maximum errors al so obviously decrease to －4.8% and 5.2% compared with －24.6% and －28.1% using LS modeling. In all,this non-linear modeling method based on standard deviation of differential optical density possesses great potential to be applied to the flue gas concentration de termination with an improved accuracy.