Bulk algorithms are used to calculate air-sea fluxes of momentum and heat (e.g. Liu et al. 1979, Fairall et al. 1996). The disadvantage is that these algorithms are iterative processes, which means that they converge slowly, when they do. In order to address these issues that are critical when processing large datasets, a statistical model of air-sea fluxes based on an artifical neural network was developed. Such a model is 10 to 100 times faster than the usual bulk algorithms. A gobal flux dataset plus data collected during three field experiments were used to show that the fluxes produced by the neural network were not biased with respect to the bulk fluxes. Moreover, our results indicate that the rms deviation between bulk fluxes and neural network estimates is only 2-4% for the momentum flux as well as the heat fluxes.