In this work, the accuracy of the Doppler beam-swinging (DBS) technique for wind measurements is studied using an imaging radar-the turbulent eddy profiler (TEP) developed by the University of Massachusetts, with data collected in summer 2003. With up to 64 independent receivers, and using coherent radar imaging (CRI), several hundred partially independent beams can be formed simultaneously within the volume defined by the transmit beam. By selecting a subset of these beams, an unprecedented number of DBS configurations with varying zenith angle, azimuth angle, and number of beams can be investigated. The angular distributions of echo power and radial velocity obtained by CRI provide a unique opportunity to validate the inherent assumption in the DBS method of homogeneity across the region defined by the beam directions. Through comparison with a reference wind field, calculated as the optimal uniform wind field derived from all CRI beams with sufficient signal-to-noise ratio (SNR), the accuracy of the wind estimates for various DBS configurations is statistically analyzed. It is shown that for a three-beam DBS configuration, although the validity of the homogeneity assumption is enhanced at smaller zenith angles, the rootmean-square (RMS) error increases because of the ill-conditioned matrix in the DBS algorithm. As expected, inhomogeneities in the wind field produce large bias for the three-beam DBS configuration for large zenith angles. An optimal zenith angle, in terms of RMS error, of approximately 9°-10° was estimated. It is further shown that RMS error can be significantly reduced by increasing the number of off-vertical beams used for the DBS processing.