Abstract: It is known that a remote sensing or an atmospheric radiative transfer equation is a non-linear Fredholm integral equation of the first kind. As a result, it is illposed,the solution is unstable,and difficulties arise in its retrieval. So it is necessary to study the method of controlling instability of retrieval and to improve the accuracy of solution. To make the solution stable, a smoothing parameter γ needs to be used as a constraint. In this paper, atmospheric profiles are retrieved by using the Newton′s non-linear iteration method. We have developed a discrepancy principle (DP) to determine γ in an objective way. An approach is formulated for achieving an optimal solution for atmospheric profiles together with γ using step by step iterations. The DP method was applied to Moderate Resolution Imaging Spectroradiometer (MODIS) infrared data. The results show that the retrieved atmospheric profiles are better than with those from the empirical method,which uses a fixed smoothing parameter.The retrieved surface skin temperature and total precipitable water vapor (TPW) are as accurate as those of the MOD07 products.