The Weibull three-parameter model is discussed for estimation of mean wind power densities. This probability density function is a generalization of a number of more conventional density functions. Using wind speed observations, it is shown that this model generally given a more reliable fit to the empirical wind speed frequency data than the density functions with one or two parameters. Wind power density estimations turn out to be strongly dependent on the hypothesized probability density function. The variation with height of the three parameters of the discussed model is investigated ; no simple height dependence can be proposed. The variability of the yearly mean wind power with time is examined for three stations in Belgium and it is found that deviations of + 20 percent of the overall mean value may occur. The geographical variation of the mean wind power density and two examples of a contour map for the generator capacity factor are presented.