The development of modern engine management systems makes ever-more stringent demands of the tools used. In future, the Hardware-in-the-Loop (HiL) simulation, used primarily for hardware and software tests to date, is also to be used for control function parameter adaptation tasks. This results in the need to provide highly precise, real-time-capable simulation models in all phases of the development process. This can be done by the use of modern methods for identification of non-linear, static and dynamic multi-variable systems, partly in conjunction with conventional physical model structures. In particular, artificial neural networks prove flexible in use in this case. This allows modelling dependent on the information available in the various phases of the engine development process. Thus, in the early phase, it is possible to develop engine models with computation results from complex engine simulation programs such as PROMO or GT Power. Methods of design of experiments (DOE) allow a high accuracy to be achieved with little modelling effort. Use of dynamic neural networks allows modelling for the non-stationary behaviour on the basis of measurements even where no confident statements are possible with complex simulation programs. This will be demonstrated by way of example of emissions.This paper represents a supplement, comprising example applications of modern, non-linear identification methods, to a treatise  which was presented at the SAE World Congress and which predominantly deals with methods of real- time modelling in early development phases.