Abstract: There are few effective ways to explore the middle atmospheric wind field directly at the altitude range of 20—60 km, and the direct sounding methods have some limitations, but the wind field could be derived from atmospheric refractive index and pressure data. From the bending angles, a large number of profiles of atmospheric refractivity, pressure and temperature are obtained with the newly launched Constellation Observing System for Meteorology Ionosphere and Climate (COSMIC)/Formosa Satellite 3(FORMOSAT-3) System. Taking full advantage of these data has a positive impact on the research of the global middle atmospheric wind field. The approach for calculating middle atmosphere zonal mean winds at the altitude range of 20—60 km is constructed according to gradient wind equations from atmospheric refractive index data, considering the characteristics and calculation methods of geostrophic wind, gradient wind and balance wind respectively, and the relationships among atmospheric refractive index, density and wind field. Following the method constructed above, the middle atmosphere zonal mean winds are calculated by the gridded refractive index data in January, April, July and October of 2007 and the latitude-height distributions of zonal mean winds are discussed. The gridded data is derived through the inverse distance weighted interpolation method. The data is compared with monthly average wind data of European Centre for Medium Range Weather Forecasts Reanalysis-interim (ERA-interim) and the Modern Era Retrospective-analysis for Research and Applications (MERRA) data sets for validation. The comparisons reveal excellent agreement, and the characteristics of calculated winds are similar with that of the reanalyzed winds. In January and July, easterly winds prevail in summer hemisphere zonal mean zona1 winds and it increase as the height increases, while in winter westerly winds prevail hemisphere zone-mean zona1 winds. The zonal wind first increases and then decreases from the high-latitude to the low-latitude regions of winter hemisphere, with the maximum in the middle-latitude regions of winter hemisphere. The root mean square deviation and the largest deviation at different heights are larger and larger along the heights, while the correlation coefficients along latitude get smaller, but it is still greater than 0.98. The root mean square deviation is about 6 m·s^-1, and the largest deviation is less than 11 m·s^-1in January and July. Spring and autumn are the transition periods, when westerly winds prevail in global, but decrease versus increasing heights in the high-latitude regions of northern hemisphere and even reverse near the top in April; westerly winds prevail in the high-latitude regions, while for some altitudes in the low-latitude regions easterly winds are dominant. The differences are not very large in April and October, with the largest deviation no more than 8 m·s^-1, indicating that deriving wind fields from the COSMIC refractive index data through gradient wind equations is an effective way.