Nitrogen oxides (NOx) emission control is a critical task of engine control as far as raw emissions and after treatment is concerned. Hardware NOx sensors are now available, but they still have several limitations, starting from their costs. Against this background, a lot of work inside both companies and in the academy has been done to develop virtual sensors for these emissions. In this paper, we propose a minimum-complexity model in the form of a piecewise affine model, and show that its performance is comparable to more complex models and better than conventional sensors if dynamic response is concerned.