The VAD (Velocity-Azimuth Display) method, which is used to retrieve wind field based on one wind lidar system and the assumption of uniform wind in the same level, has been widely used in the industry. To guarantee the accuracy, the azimuthal range of lidar and number of scanning should meet some requirements, which affects measurement efficiency. Based on this, a new wind retrieval method of VAD used for Doppler lidar based on gradient descent algorithm was developed. The gradient descent algorithm was used to replace present Fourier series expansion for the solution of VAD. The convergence and calculation speed were improved by analyzing the factors that affect algorithm convergence and determining the iteration step length and the number of iterations, respectively. In order to demonstrate this modified VAD method, the wind data of coherent Doppler lidar was compared with standard wind cup anemometer (IEC 61400-12-1). The correlation coefficients of the 10 min- averaged wind speed and direction were up to 0.99 in the case that the azimuth range and the number of radial velocity were 60° and 7, with wind speed standard deviation and bias of 0.52 m/s and 0.02 m/s, and with wind direction standard deviation and bias of 5.1° and 3.6°. The results prove that this modified VAD method could improve Doppler measurement efficiency and applicability, and guarantee its accuracy at the same time, which improves the system capability for the monitoring of dynamic complex wind field.