This paper presents a model for on-line NOx estimation. The method uses both, low frequency components and high frequency components of in-cylinder pressure signal: it harnesses in-cylinder pressure resonance to estimate the trapped mass, and based on this measurement, a NOx model is adapted to estimate NOx emissions cycle by cycle. In addition of the in-cylinder pressure signal, the procedure only requires from lambda and air mass flow to estimate NOx, so it can give a direct estimation of NOx or improve transient response and aging of current NOx sensors. The method was validated on a CI engine with high pressure EGR loop under steady and transient conditions showing errors below 10% and cycle by cycle time response. (C) 2016 Elsevier Ltd. All rights reserved.