A 2-dimensional variational method is used to blend the satellite scatterometer measured (QuikSCAT) and regional mesoscale atmospheric model simulated (COAMPS) surface vector winds for coastal central California. The approach is distinct from existing methods in that it considers errors from both measurements and models. When compared with independent in situ observations, the blended wind product shows consistently higher correlation and smaller RMS errors than QuikSCAT or COAMPS winds. The proposed algorithm can be implemented over any part of the world ocean. It should be a valuable tool for describing small-scale atmospheric processes in coastal zones and for forcing high-resolution coastal ocean models.