The capability of producing accurate and reliable wind forecasts has a significant impact on offshore navigation, both for on board safety and for taking tactical decisions during sailing yacht races. In this paper we describe a prediction model based on Artificial Neural Networks (ANNs), which have been proven to be very powerful in simulating non-linear systems. Their main feature is the ability of learning adapting themself to the evolution of the environment. A time-series approach is used, and the forecast is based on past wind data. The proposed algorithm takes as input the wind speed registered at previous instants and predicts the future wind speed for a short period. The algorithm is extended in order to compute a distribution of probability for the possible future evolution of the wind scenario. The interesting feature of this method is that only past wind data is used to obtain the forecast and its probability. Preliminary studies have shown encouraging results, allowing very accurate short-term predictions both for artificially constructed signals and real wind data. Different architectures for the Artificial Neural Networks are discussed and the characteristics of the optimum algorithm, which maximise the accuracy of the prediction, are presented. The present algorithm is under continuous development and additional enhancements will be presented in the full paper.