To support very long baseline interferometric experiments, the Jet Propulsion Laboratory has developed a system for estimating atmospheric water vapor path delay. The system consists of dual microwave radiometers, one operating at 20.7 GHz and the other at 31.4 GHz. The measured atmospheric brightness temperatures at these two frequencies yield the estimate of the precipitable water present in both vapor and droplets. To determine the accuracy of the system, a series of observations were undertaken comparing the outputs of two water vapor radiometers with the phase variation observed with two connected elements of the very large array (VLA). The results show that (1) water vapor fluctuations dominate the residual VLA phase and (2) the microwave radiometers can measure and correct these effects. The rms phase error after correction is typically 15° at a wavelength of 6 cm, corresponding to an uncertainty in the path delay of 0.25 cm. The residual uncertainty is consistent with the stability of the microwave radiometer but is still considerably larger than the stability of the VLA. The technique is less successful under conditions of heavy cloud.