With differential optical absorption spectroscopy (DOAS), trace gases in the atmosphere can be measured and its concentrations can be retrieved based on least squares regression. Under complicated atmospheric conditions, there are outliers and the error distribution is not normal in DOAS differential spectra, which resulted in misestimate of least-squares regression. The retrieving model of robust regression based on M-estimator was developed to evaluate concentrations of trace gases in DOAS system. The evaluation procedure and effects using M-estimator robust regression were studied. The normal spectrum and abnormal spectrum were retrieved basing on two methods. Experimental results show that reliability is improved with method of robust regression in DOAS evaluation.