In order to continue cost-optimization of modern large wind turbines, it is important to continuously increase the knowledge of wind field parameters relevant to design loads. This paper presents a general statistical model that offers site-specific prediction of the probability density function (PDF) of turbulence driven short-term extreme wind shear events, conditioned on the mean wind speed, for an arbitrary recurrence period. The model is based on an asymptotic expansion, and only a few and easily accessible parameters are needed as input. The model of the extreme PDF is supplemented by a model that, on a statistically consistent basis, describes the most likely spatial shape of an extreme wind shear event. Predictions from the model have been compared with results from an extreme value data analysis, based on a large number of full-scale measurements recorded with a high sampling rate. The measurements have been extracted from "Da tabase on Wind Characteristics" (http://www.winddata.com/), and they refer to a site characterized by a flat homogeneous terrain. The comparison has been conducted for three different mean wind speeds in the range of 15-19 m/s, and model predictions and experimental results are consistent, given the inevitable uncertainties associated with the model as well as with the extreme value data analysis.