For control-oriented modeling of engine out NOx emissions a vast literature exists. The methods applied to derive suitable models range from basic physical considerations in which only some parameters are adjusted or optimized to purely data driven approaches. The first ones generally yield more insight into the underlying principles and help to improve understanding. Examples for data based approaches are black box models. In this case a model is derived from the data alone without the inclusion of any system knowledge. The black box models in general provide more degrees of freedom but their quality heavily relies on the data utilized for model identification. Thus, the choice of the model structure strongly depends on the purpose the model is used for. In this paper the focus lies on black box NOx-models which are capable to run in real-time on a standard production engine control unit (ECU), thus only standard ECU signals are considered as model inputs. The nominal model is extended with additional inputs ? intake air humidity and the engine coolant temperature ? which affect the total NOx emissions and are captured by a multiplicative correction factor.