A molecular potential energy surface (PES) is fitted by a neural network (NN) using some Hartree-Fock data as inputs. The newly interpolated PES is then tested by running some classical dynamics using the velocity-verlet algorithm.