We present evidence that modeling troposphere delay gradients in precise Global Positioning System (GPS) geodesy improves the accuracy and precision of the estimated quantities, and that the estimated gradients resemble real atmospheric moisture gradients observed with a water vapor radiometer (WVR). Using a low elevation angle cutoff, combined with a model of the atmospheric delay gradient as a random walk process leads to 19.5% and 15% average improvement in radial and horizontal site position repeatabilities, respectively, relative to a current state-of-the-art estimation strategy that does not model horizontal gradients and imposes high elevation angle cutoff. The agreement between estimated values of zenith wet delay from collocated GPS receivers and WVRs was improved by at least 25%. Merely lowering the elevation angle cutoff improves the repeatability of the radial component of the site's position vector but tends to degrade the repeatability of the horizontal components of the position vector if troposphere gradients are not properly modeled. The estimates of wet delay gradients from a collocated GPS receiver and a WVR at Onsala, Sweden, seem to be correlated over timescales as short as 15 min. The agreement in azimuth between the GPS-based and the WVR-based gradients was at the 10° level, for significant gradients. The GPS was found to under-estimate the magnitude of the gradients by about 60% relative to the WVR-based gradients. The ability to sense atmospheric moisture gradients from a single GPS receiver increases the useful information content from networks of GPS receivers by providing additional spatial information for weather forecasting applications.