Abstract The statistical study of wind speed measurements on a site makes it possible to determine a distribution law, needed to assess the available or recoverable wind energy potential. The classical approach consists in assimilating the distribution law to standard models, for example Weibull or Rayleigh, and in determining the parameters of the model so that it gets closest to the discrete law obtained by statistically treating the wind speed measurements. The Weibull model is the most used one and provides good results. However, the accurate determination of the wind speed distribution law constitutes a major problem. Multi Layer Perceptron type artificial neural networks, highly effective in function approximation problems, are used here for the approximation of the wind speed distribution law. The site energy characteristics have been determined by means of the neural approach and compared with those obtained by the classical method. The results show that the distribution law achieved by the neural model provides assessments closer to the discrete distribution than the Weibull model. This approach has enabled the wind energy potential on the Dakar site to be determined in a more accurate way. The models are also used to assess the amount of energy the wind generator WES18 of 80 kW power, set up at 10 m and 40 m above the ground, would produce annually.