Probability distributions of the three wind velocity components measured at a height of 26m near the coast in Keelung, Taiwan were studied using statistical models found in the literature. Both the linear Gaussian distribution and a weakly nonlinear model were used. It is shown here that, for this specific site, the model based on the Gram-Charlier type A series expansions can be used to model all the three fluctuating velocity components more satisfactorily than the Gaussian model.