The GSETT-3 experiment was conducted for more than two years and the International Data Center (IDC) reported more than 20,000 events for 1995 in the Reviewed Event Bulletins (REB). We have used this data base to investigate the location bias of arrays and selected three-component stations in the Alpha network, for which the recorded data is automatically processed to determine earthquake locations in near-real time. The Alpha network, which consists of 14 arrays, provides estimates of event azimuths determined by f-k processing. We selected 19 three-component stations that were operated all or most of 1995 and for which event azimuth was estimated by polarization analysis. Only events with epicentral distances of less than 100° were considered in order to exclude phases with steep incidence angles, associated with unreliable azimuth estimates. Teleseismic P phases and the major regional and local phases were examined. Only data for which a signal-to-noise ratio was given were included in this analysis. The signal-to-noise ratios were used as a weighting factor for the statistical parameters and the graphical representation of the azimuth residuals. The preliminary analysis of azimuth residuals identified some arrays that give significantly erroneous azimuths; in some cases, the error is systematic, in other cases, it is random. One array with systematic azimuthal bias is CMAR. This array shows positive residuals for event paths from the south, while phases for events with paths from the north show negative residuals. Similar patterns, which strongly suggest azimuthal bias, are observed for PDAR, TXAR, and SPITS. In the case of MJAR, the large azimuthal errors observed particularly for regional and local distances are not a function of direction. The scatter in the azimuth residuals is much larger for 3-C stations than for the arrays. Clear systematic errors were identified for SCHQ and CPUP, which show constant offsets, and at VNDA, which has negative residuals for events to the north and positive residuals for events from the south. The most significant azimuth bias was observed for PDY, where the azimuth residuals change from about −90° to more than +90° at backazimuths near 120°, meaning that all rays appear from the northwest.