A statistical method to correct for the limb effect in off-nadir Atmospheric Infrared Sounder (AIRS) channel radiances is described, using the channel radiance itself and principal components (PCs) of the other channel radiances to account for the multicollinearity. A method of selecting an optimal set of predictors is proposed and demonstrated for one- and two-PC predictors. Validation results with a subset of AIRS channels in the spectral region 649–2664 cm−1 show that the mean nadir-corrected brightness temperature (BT) is largely independent of scan angle. More than 66% of the channels have a root-mean-square (rms) bias less than 0.10 K after nadir correction. Limb effect on the standard deviation (SD) of BT is discernible at larger scan angles, mainly for the atmospheric windows and the water vapor channels around 6.7 μm. After nadir correction, nearly all atmospheric window channels unaffected by solar glint and more than 76% of water vapor channels examined have BT SDs brought closer to nadir values. For the window channels affected by solar glint (wavenumber > 2490 cm−1), BT SDs at the scan angles with the strongest impact from solar reflection were improved on average by more than 0.6 K after nadir correction.