Highspectralresolution Atmospheric Infrared Sounder (AIRS) data can be used to retrieve the small scale vertical structure of air temperature, which provid ed a more accurate and fine initial field for the numerical forecasting and the largescale weather analysis. In the previous studies, eigenvector regression algorithm in the IMAPP (International MODIS/AIRS Preprocessing Package) was ofte n used to process the data. Because of its simplicity, the inversion precision w as limited. Applying an artificial neural network to retrieve the clear sky atm ospheric temperature profiles from AIRS simulation radiation data and comparing with the eigenvector regression algorithm, the results indicate that the neural network consumed a same time as the eigenvector regression algorithm, but it red uced the atmospheric inversion error and made improvements in temperature measur ements at various levels. 