Analysis of Global Positioning System (GPS) data from two sites separated by a horizontal distance of only ~2.2 m yielded phase residuals exhibiting a systematic elevation angle dependence. One of the two GPS antennas was mounted on an ~1-m-high concrete pillar, and the other was mounted on a standard wooden tripod. We performed elevation angle cutoff tests with these data and established that the estimate of the vertical coordinate of site position was sensitive to the minimum elevation angle (elevation cutoff) of the data analyzed. For example, the estimate of the vertical coordinate of site position changed by 9.7¿0.8 mm when the minimum elevation angle was increased from 10¿ to 25¿. We performed simulations based on a simple (ray tracing) multipath model with a single horizontal reflector which demonstrated that the results from the elevation angle cutoff tests and the pattern of the residuals versus elevation angle could be qualitatively reproduced if the reflector were located 0.1--0.2 m beneath the antenna phase center. We therefore hypothesized that the elevation-angle-dependent error was caused by scattering from the horizontal surface of the pillar, located a distance of ~0.2 m beneath the antenna phase center. We tested this hypothesis by placing microwave absorbing material between the antenna and the pillar in a number of configurations and by analyzing the changes in apparent position of the antenna. The results indicate that (1) the horizontal surface of the pillar is indeed the main scatterer, (2) both the concrete and the metal plate embedded in the pillar are significant sources of scattering, and (3) the scattering can be reduced greatly by the use of microwave absorbing materials. These results have significant implications for the accuracy of global GPS geodetic tracking networks which use pillar-antenna configurations identical or similar to the one used for this study at the Westford WFRD GPS site. ¿ American Geophysical Union 1995