A method of estimating vapour liquid equilibrium (VLE) using NMR chemical shift data has been proposed by Xu et al. (2012). This method is based on the concept that the average local composition around each species is determined by the thermodynamics of the system, and also determines the screening of the NMR active groups within that molecule, and so their NMR chemical shifts. Xu et al. (2012)'s method has been replicated and verified; results are confirmed to be accurate for alcohol+hydrocarbon mixtures, giving VLE predictions of comparable accuracy to the UNIFAC, generally considered the best predictive activity coefficient model available. However, for more strongly non-ideal mixtures, the method becomes less reliable, giving significantly less accurate predictions of total pressure than UNIFAC. Several causes for this are identified. The model proposed by Xu et al. (2012) is unable to fit minima or maxima in chemical shifts, which are observed experimentally in some binary mixtures. Different NMR resonances within the same molecule lead to different predictions of VLE, clearly an un-physical result. The thermodynamics of strongly non-ideal mixtures are determined by more complex interactions than a simple description of average local composition around each component in the mixtures, for example strong and directional hydrogen bonds. Different groups within the same molecule may have different local compositions in their immediate vicinity; for example in the case of alcohol+water mixtures, one would expect a clustering of water molecules around the hydroxyl group but not the aliphatic group. Hence, the concept of a simple local composition model is not valid for these more complex cases, and it is therefore not surprising that a model based on this simple concept is often not effective in predicting VLE. (C) 2014 Elsevier Ltd. All rights reserved.