Strong winds may be biased in atmospheric models. Here the ECMWF coupled wave-atmosphere model is used (1) to evaluate strong winds against observations, (2) to test how alternative wind stress parameterizations could lead to a more accurate model. For the period of storms Kaat and Lilli (23 to 27 January 2014), we compared simulated winds with in-situ – moored buoys and platforms - and satellite observations available from the North Atlantic. Five wind stress parameterizations were evaluated. The first result is that moderate simulated winds (5-20 m s-1) match with all observations. However, for strong winds (above 20 m s-1), mean differences appear, as much as -7 m s-1 at 30 m s-1. Significant differences also exist between observations, with buoys and ASCAT-KNMI generally showing lower wind speeds than the platforms and other remote sensing data used in this study (AMSR2, ASCAT-RSS, WindSat, SMOS and JASON-2). It is difficult to conclude which dataset should be used as a reference. Even so, buoy and ASCAT-KNMI winds are likely to underestimate the real wind speed. The second result is that common wave-age dependent parameterizations produce unrealistic drags and are not appropriate for coupling, whereas a newly empirically-adjusted Charnock parameterization leads to higher winds compared to the default ECMWF parameterization. This proposed new parameterization may lead to more accurate results in an operational context.