The spectrally resolved infrared radiances observed by AIRS provide a strict and insightful test for general circulation models (GCMs). We compare the clear- and total- sky spectra simulated from the Geophysical Fluid Dynamics Laboratory GCM using a high resolution radiation code with the AIRS observations. After ensuring consistency in the sampling of the observed and simulated spectra and a proper representation of clouds in the radiance simulation, the observed and simulated global-mean radiances are shown to agree to within 2 K in the window region. Radiance discrepancies in the water vapor v2 (1300–1650 cm⁻¹) and carbon dioxide v2 (650–720 cm⁻¹) bands are consistent with the model biases in atmospheric temperature and water vapor. The existence of radiance biases of opposite signs in different spectral regions suggests that a seemingly good agreement of the model's broadband longwave flux with observations may be due to a fortuitous cancellation of spectral errors. Moreover, an examination of the diurnal difference spectrum indicates pronounced biases in the model-simulated diurnal hydrologic cycle over the tropical oceans, a feature seen to occur in other GCMs as well.