Polynomial translation method for translating Gaussian random variables to non-Gaussian random variables has been proposed for more than half centuries. Application of this method has also been utilized to many aspects. However, the approximation of parameters of the polynomial form cannot provide good estimation when random variables are given large skewness and kurtosis. In this research, the approximation method used for the parameter estimation is investigated for wider ranges of given skewness and kurtosis, which are observed from the 10 minute mean wind speed samples of 155 sites from 1961 to 2002. The proposed simulation process of the estimation of annual maximum wind speeds is then modified based on the investigation results. For those extreme cases with large observed skewness or kurtosis, the modified simulation process can provide fairly good estimation results as normal cases.