This paper aims to improve the estimation of turbulent fluxes derived from satellite observations, and to assess the comparison results obtained from in-situ and atmospheric model analysis or re-analysis data. Global analysis of surface wind and latent heat flux are performed using merged values of surface variables obtained from satellite radar and radiometer data. Two scatterometers, the European Remote Sensing Satellite 2 (ERS-2) and the NASA scatterometer (NSCAT)), and four Special Sensor Microwave/ Imager' s(SSM/I's: F10, F11, F13 and F14) are used. For winds, daily, weekly, and monthly fields are calculated, while for latent heat flux, only weekly and monthly fields are estimated. Intercomparisons involving satellite and in situ data (from buoys and the COADS ship data ) are performed over a wide range of oceanic conditions during the period October 1996 - September 1997, to determine the accuracy and the errors of the wind and latent heat flux estimates. For instance, in the tropical Pacific area, the rms difference between satellite and buoys (TAO) are about of 1.5m/s for wind speed, and 27W/m^2 for latent heat flux. The comparisons with COADS data provides rms values of about 2m/s, and 39W/m^2.The most significant spatially coherent latent heat flux errors are related to the estimation of the specific air humidity (qa) from SSM/I brightness temperatures. A new inversion model is proposed and used in qa calculation. At global scales, the accuracy of satellite wind and heat flux fields is assessed through a comparisons with ECMWF analysis and NCEP re-analysis. In terms of surface winds, atmospheric model winds are underestimated with respect to satellite values. The latent heat flux comparisons show regional and seasonal bias. In north Atlantic and during winter, the bias between ECMWF and satellite weekly latent heat flux estimates exceeds 70W/m^2. To assess the quality of numerical analysis and satellite turbulent flux comparisons, a triplet collocated data set derived from numerical model, in-situ and satellite data is used. Statistical error analysis are then performed.