Model-based optimisation is gaining more and more ground in automotive engineering. The underlying models have to be smooth, fast, quantitatively accurate, and need to provide plausible extrapolation. A previously published model for the NOx emissions of a Diesel engine incorporates all these requirements but relies on the measured in-cylinder pressure. In this paper, that model is embedded in a mean-value model for the air path. Thus, the NOx emissions can be predicted based on the control signals only. Physics-based models are developed for the interfacing processes such as the cylinder charge, its composition, the compression, and the ignition delay. The accuracy of the original model is preserved, while the execution time of the extended NOx model is found to be only a fraction of that of the mean-value engine model. An experimental validation on an actual transient driving cycle highlights the advantages of the procedure presented. (C) 2013 Elsevier Ltd. All rights reserved.