NOX emissions are one of the major limiting factors of modern diesel engine technology; they heavily influence, directly or indirectly, both engine and after-treatment design, cost, complexity and reliability; they are also linked in an important trade-off with CO₂ emissions and therefore fuel consumption. It is paramount for OEMs (Original Equipment Manufacturers) to exploit more sophisticated techniques for modeling the formation of NOX to reduce costs and increase their ability to meet the legislative requirements for both CO₂ and NOX. Many existing simulation models predict NOX simply by interpolating steady state engine maps with limited ability to efficiently capture the effects of engine warm up, speed-load transients and air system dynamics. For conventional powertrains running on light cycles this might still be acceptable, but it becomes inadequate when applied to fast and deep transients across unconventional speed and load patterns. On the other side of the spectrum, extremely sophisticated, fully predictive combustion models are just too complex to be attractive as a development tool.The objective of this paper is to describe a semi-empiric model based approach for dynamic NOX emission modeling that is being developed by Prodrive as part of the FHSPV (Flywheel Hybrid System for Premium Vehicle - www.FHSPV.org) consortium. The required measured data are kept to a minimum and are still primarily based on stationary engine maps recorded on test bed. These maps determine the steady state component of the NOX prediction, the dynamic part being calculated based on key engine parameters. From onboard sensor data the model calculates in-cylinder conditions at Inlet Valve Closing (IVC); based on an average wall temperature it then calculates the conditions at the start of the compression stroke and, based on a Wiebe heat release model, it determines the degree-by-degree profiles for pressure and temperature. The model uses a simplified Zeldovich mechanism to calculate NOX. The result is used to validate the model against the measured steady-state engine maps. Attention will be given to the correlation process that enables the shift from an angle-based to a time-based domain as this is a key aspect of this approach and one which can be finely tuned to vary the fidelity of the model based on the requirements of the end-user.In conclusion the paper demonstrates the predictive performance of the model in relation to transient events within the NEDC (New European Drive Cycle). It also lists key advantages and suggests the next steps to address its limitations.