The quest for a single universally applicable probability distribution for wind speed is reviewed and shown to be largely futile due to the "mixed" nature of wind climates. The physical justification, derivation and development of the OEN mixture model for the joint distribution of wind speed and direction is described. An automatic, unsupervised method for fitting wind observations is proposed, implemented as R scripts and applied to wind records from 128 WMO reporting stations, distributed globally and in every Koppen climate class to give global coverage. The effect of the standard 1 kn speed and 10 degrees direction reporting intervals on the rms error and the coefficient of determination, R-2, is assessed by bootstrapping and is used to set the target goodness-of-fit criteria. The OEN mixture model is shown to meet the target criteria at all 128 stations, demonstrating its global potential as a physically realistic model for joint and marginal wind speed and direction distributions. R scripts to reproduce the results and apply the method to other sites are freely available for download and use from the Mendeley archive.