AIRS(Atmospheric Infrared Sounder), the first high spectral atmospheric infrared detector, started a new era of satellite sounding atmosphere. The samples, composed of the radiosonde observations and the AIRS brightness temperature value simulated by the SARTA (StandAlone Radiative Transfer Algorithm)v1.05 forward mode, were inversed to the atmospheric humidity profiles using neural network. The results show that, compared with the result of eigenvector statistics, the neural network inversion is of higher precision, and the humidity profiles obtained are closer to the true profiles. AIRS displays the fine structure of the atmosphere because of its high spectral resolution (high vertical resolution).Neural network has a strong nonliner processing capability in the issue that the inversion of the atmospheric humidity profiles is based on high spectral data.