To analyze the error produced when the atmospheric parameters are inverted by using the classical physical method, the influence of channel characteristics, background error, observation error and instrument noise etc on inversion accuracy is analyzed by using the Bayesian principle during the inversion of atmospheric parameters. An index for quantitatively measuring the contribution of channels to the inversion accuracy of atmospheric parameters is proposed. Taking this index as a criterion, a channel selection scheme is designed and demonstrated when the atmospheric temperature profile is inverted by using high spectral resolution data. 300 channels which have the greatest contribution index are selected from 8461 channels of the second-generation Infrared Atmospheric Sounding Interferometer (IASI) launched in 2012 and are compared with the application channels published by AAPP. The result shows that the proposed channel selection method is effective and feasible for the inversion of atmospheric temperature profile.