Due to the underdetermined nature of the inverse problem for SAR wind, ancillary information is routinely used. Up to now one of the more mature approaches is certainly to use radar cross section in combination with wind information from a numerical model into a Bayesian scheme. In this study, we investigate the potential of using the Doppler shift estimated from the SAR as new information to include in the Bayesian scheme. The output of this new wind inversion algorithm is the optimal wind vector which minimizes the total deviation from the input information, quantified by a cost- function derived from the Bayesian theorem of probability. Validation of the algorithm against in situ wind measurements shows a significant decrease of error, compared with a simpler algorithm using only sigma0 and model wind direction to invert the CMOD4 geophysical model function. For situations such as fronts and cyclones, the new algorithm is seen to give a more realistic output wind field, thanks to the Doppler Centroid Anomaly which precisely locates sharp changes of wind direction. Moreover an intensive quantitative validation against ASCA T L2 wind products (spatial resolution is 12.5 km) give reasonable statistical results. Quantitative validation against in situ measurements shows however that the Bayesian algorithm performs slightly better when the Doppler part is omitted from the cost function. The interpretation is that whereas the Doppler shift is very useful to detect sharp wind direction changes, the absolute calibration of this radar quantity can certainly be improved. As a result, when the numerical model is very consistent with the radar cross section, the Doppler can add noise in the inversion scheme. On the contrary, when the numerical model is not precise enough (in cases of rapidly changing meteorological situations such as atmospheric fronts) the Doppler information can be useful.