Measurements with an airborne microwave radiometer are used to validate absorption models in clear air at frequencies used in satellite microwave humidity sounders, such as AMSU-B and MHS. The data set includes 33 profiles of radiometric and in situ observations ranging from the tropics to the high arctic. In cold, dry conditions the MPM and Rosenkranz models were found to underestimate the observed downwelling radiances near the surface in window channels dominated by the water vapor continuum. The models' absorption coefficient is calculated and found to be strongly deficient compared to observations in layers of low specific humidity. For medium humidity, the models' negative bias persists at 89 GHz and becomes statistically significant, but reduces at higher frequencies. Previously, the radiometer calibration had been adjusted to fit modeled radiances during high-level flight; this approach has been revised in the light of these findings and a tip curve method applied. Possible spectroscopic reasons for the models' deficits are suggested, including the underestimation of continuum terms or the extrapolation of oxygen line coupling coefficients to low temperatures. The impacts for operational use of radiative transfer models are discussed.