Potential Energy Surfaces (PES) for the ethanol/Au(1 1 1) interface are mapped by Neural Networks (NNs). Interaction energies, calculated from Density Functional Theory (DFT), for the adsorption of the ethanol on Au(1 1 1) surfaces are used to train Ensembles of Feed-Forward Neural Networks (EnsFFNNs). The distance of the ethanol molecule to the surface, two angles describing the molecular orientation relatively to the surface, and three binary descriptors encoding the gold adsorption sites, are the input to the NNs. The training sets contain energy values at different distances, for seven molecular orientations and three adsorption sites. The models are assessed by: (a) internal cross validation; (b) Leave-One-Out procedure (LOO); and (c) external test sets corresponding to orientations not used in the training procedure. The results are compared with the ones obtained from an analytical force field recently proposed by some of us to match the DFT data. It is shown that NNs can be trained to map PES with a similar or better accuracy than analytical representations. This is a relevant point, particularly in simulations by Monte Carlo (MC) or Molecular Dynamics (MD), which require an extensive screening of the interaction sites at the interface, turning the development of analytical functions a non-trivial task as the complexity of the systems increases. (C) 2008 Elsevier B.V. All rights reserved.