The Velocity Azimuth Display (VAD) techniques are often used to estimate a wind profile over a Doppler radar. The method obtain wind profiles by using radar radial velocity observation. Difficulties in this technique occur when the data are contaminated by velocity ambiguities, large noises and when there is a restricted azimuth range where good data exists. The method proposed in this paper using gradients of radial velocity as input to derive wind profiles, so they are called GVAD method.The VAD method is tested on idealized data first to examine the sensitivity to different type of errors in radial velocity data. It is found that RMS errors in the mean wind profiles retrieved using both the VAD and GVAD methods are not sensitive to random errors in the radial velocities, even with large amplitude. However, the VAD method is very sensitive to system errors caused by velocity ambiguities. The experiment indicates that even if only 3% of a full volume scan of radial wind data are contaminated by alias errors, relative RMS errors in the mean wind profile retrieved by VAD can reach 50%; In the contrast, GVAD is very robust to resist velocity alias errors. Tests of GVAD method on a set of WSR-88D collected during 3 May, 1999 Oklahoma severe storm cases show that GVAD method has the ability to obtain wind profiles reasonable even when raw data contain big errors caused by velocity ambiguities and large random noises.