This paper develops the models that can be used in different flight scenarios to retrieve the wind speed using the width of the normalized delay waveform (NDW). First, the factors that influence the NDW width, including the wind speed, wind direction, flight height, and elevation angle, are analyzed. The contribution of each independent variable to the regression is explored. The results show that the wind speed, flight height, and elevation angle contribute more significantly than the wind direction so the wind direction can be ignored in the model. The multiple regression, in which the function terms of NWD width, flight height, and elevation angle above are taken as independent variables and wind speed is taken as the dependent variable, is proposed to develop the model of retrieving wind speed. Through the simulation, a root-mean-square error (RMSE) over 3 m/s can be obtained. In order to improve the retrieval performance, a Back-Propagation (BP) network is trained as an alternative to the analytical models above. Better performance is achieved with an RMSE less than 2.5 m/s under the same conditions with the analytical model. The errors of retrieved wind speed are analyzed. The conclusions are that: 1) both analytical model and BP network have inherent regression biases, especially for high wind speed from 15 to 20m/s so that at wind speeds higher than 16 m/s, the tendency for retrieval accuracy to rapidly become worse appears and 2) the number of incoherent averaging should be over 1000 to reduce the impact of thermal and speckle noise. At the end of the paper, airborne data are processed to retrieve wind speed utilizing proposed models and the matching method to compare in-situ wind speed from the National Centers for Environmental Prediction (NCEP) Climate Forecast System (CFS). By comparing the retrieval results, the proposed methods could obtain the accuracy with the same level of matching method.INDEX TERMS Wind speed, reflected GNSS signal, normalized delay waveform (NDW), NDW width, multiple regression, neural network.