Abstract Computational fluid dynamic (CFD) simulations are an essential tool in the design of an aerodynamic body. These simulations tend to be computationally intensive and in the case of multi-disciplinary design optimisation (MDO) where many design iterations may be required, minimising this computation time is extremely important. In this paper, a variety of Artificial Neural Network (ANN) architectures are investigated to determine whether they could be substituted for the CFD, with the intention of decreasing the computation cost without compromising the accuracy of the results. The ANN was trained on aerofoil data which was simulated using the Stanford University Unstructured (SU2) solver. The training set comprised of NACA aerofoils varying in physical parameters and angle of attack. The goal of the ANN was to predict the corresponding lift (CL) and drag (CD) coefficients given these parameters as inputs. A Feed-Forward Back- Propagation (FFBP) network was employed for this using the TensorFlow open-source library. Many ANNs were trained covering a wide range of architectures. The best network performed with an average CL error of 1.82% and CD at 9.76% when tested against new simulations. The network achieved this with negligible losses in accuracy when compared to wind tunnel data.