A wavelet based methodology has been developed to retrieve the wind field over the sea surface from SAR images without any a-priori information of the wind direction. The method relies on the ability of the two-dimensional continuous wavelet technique to detect the spatial structure of the Marine Atmospheric Boundary Layer (MABL). The analysis of the geometric and radiometric characteristics of the backscatter structures defines the wind direction allowing the computation of the wind speed through the available algorithms. Twenty SAR images (Envisat ASAR Wide Swath and ERS) over the Mediterranean Sea have been analysed, and the results compared both with the NASA QuikSCAT satellite and the ECMWF analysis wind fields. These images cover a wide range of meteorological conditions, from low (2 m/s) to moderate winds (12 m/s), presenting many kinds of signature, i. e. wind cells, atmospheric gravity waves, convective structures and radiometric flatness. The main difference of this method with respect to those already proposed is that the wind direction is not supposed a priori, but derived only from the SAR image signature. The local scale structure of the wind field is thus retrieved with a resolution finer than that of all the existing other sources (atmospheric models and satellite scatterometers).A statistical comparison of the SAR derived wind speed and direction has been performed against QuikSCAT satellite winds (for Envisat images) or ECMWF analysis winds. Results for Envisat ASAR indicate a good score (about 75 %) in detecting the wind direction.