Collocated measurements of opacity (from water vapor radiometer brightness temperatures) and wet path delay (from ground-based tracking of global positioning satellites) are used to constrain the model of atmospheric water vapor absorption in the 20-32 GHz band. A differential approach is presented in which the slope of opacity-versus-wet delay data is used as the absorption model constraint. This technique minimizes the effects of radiometric calibration errors and oxygen model uncertainties in the derivation of a best-fit vapor absorption model. A total of approximately five months of data was obtained from two experiment sites. At the Cloud and Radiation Testbed (CART) site near Lamont, Oklahoma, three independent water vapor radiometers (WVRs) provided near-continuous opacity measurements over the interval July-September 1998. At the NASA/Goldstone tracking station in the California desert two WVRs; obtained opacity data over the September-October 1997 interval. At both sites a Global Positioning Satellite (GPS) receiver and surface barometer obtained the data required for deriving the zenith wet delays over the same time frames. Measured values of the opacity-versus-wet delay slope parameter were obtained at four WVR frequencies (20.7, 22.2, 23.8, and 31.4 GHz) and compared with predictions of four candidate absorption models referenced in the literature. With one exception, all three models provide agreement within 5% of the opacity-versus-wet delay slope measurements at all WVR frequencies at both sites. One model provides agreement for all channels at both sites to the 2-3% level. This absorption model accuracy level represents a significant improvement over that attainable using radiosondes.