A selective dynamical downscaling method is developed to obtain extreme‐wind atlases for large areas. The method is general, efficient and flexible. The method consists of three steps: (i) identifying storm episodes for a particular area, (ii) downscaling of the storms using mesoscale modelling and (iii) post‐processing. The post‐processing generalizes the winds from the mesoscale modelling to standard conditions, i.e. 10‐m height over a homogeneous surface with roughness length of 5 cm. The generalized winds are then used to calculate the 50‐year wind using the annual maximum method for each mesoscale grid point. The generalization of the mesoscale winds through the post‐processing provides a framework for data validation and for applying further the mesoscale extreme winds at specific places using microscale modelling. The results are compared with measurements from two areas with different types of extreme‐wind climates, and the results are promising. Copyright © 2012 John Wiley & Sons, Ltd.