Abstract: Atmospheric Infrared Sounder（AIRS）is on board the EOS-2/ Aqua satellite, launched on May 4, 2002. It is now an advanced atmospheric remote sensing instrument with a super high spectral resolution and a global detection coverage in the world. Study on AIRS data processing is of significance in both science and operational application. In the retrieval of atmospheric temperature from the AIRS observations, the eigenvector method is used, and collocated AIRS brightness temperature and corresponding T213 temperature analyses fields are used to compute the regression coefficients. The study includes the following aspects: selection of spectral channels, impacts of surface emissivity and topography, determination of additional predictors, and comparison of retrieval results. Research results indicate that: (1) it is very essential to rationally choose the channels when retrieving atmospheric temperature with the high spectral resolution AIRS data; (2) AIRS data can significant improve the retrieval accuracy of temperature in mid-upper troposphere and at the tropopause, however in the lower troposphere, especially near the surface, the accuracy is not satisfactory; (3) different surface emissivity and topography will affect retrieval result; (4) factors reflecting surface radiation nature, such as surface temperature, surface emissivity, and topography, are selected as additional predictors, which improves the retrieval accuracy near the surface; (5) AIRS retrieval accuracy in this paper is consistent with the AIRS team's.