This paper studies the application of Kalman filtering as a post-processing method in numerical predictions of wind speed. Two limited-area atmospheric models have been employed, with different options/capabilities of horizontal resolution, to provide wind speed forecasts. The application of Kalman filter to these data leads to the elimination of any possible systematic errors, even in the lower resolution cases, contributing further to the significant reduction of the required CPU time. The potential of this method in wind power applications is also exploited. In particular, in the case of wind power prediction, the results obtained showed a remarkable improvement in the model forecasting skill.