Differential optical absorption spectroscopy(DOAS) has become a widely used method to measure trace gases in the atmosphere. Their concentrations are retrieved by a numerical analysis of the atmospheric absorption spectra. But in the process of application, it is found the error is a bit larger and results are not steady since the limited bandwidth for each analysis subject to system hardware. A new procedure was developed, based on the Madaline artificial neural network theory to expand the spectra bandwidth. A good result was obtaind with this method according to the comparison of the different test results.