Future pollutant emissions legislations are expected to be increasingly stringent. To reduce Nitrogen Oxides (NOx) emissions produced by Diesel engines, advanced combustion technologies - like Low Temperature Combustion (LTC) -, vehicle hybridization and NOx after-treatment systems - such as Selective Catalytic Reduction (SCR) systems - can be considered, leading to a growing demand for NOx models.In this paper, we present a state-of-art of the different existing NOx models, from the black-boxes to the three-dimensional Computational Fluid Dynamics (CFD) codes. A way to classify these models is proposed. The paper also introduces the current applications for each subgroup of models.Then, a black-box and two grey-box NOx models are studied regarding their accuracy and their sensitivity to model inputs. These models are validated for two Diesel engines on steady-state operating points as well as on transient operations. The semi-physical models accurately predict NOx emissions. The static map shows a good match with experimental data for the operating conditions resulting from the engine calibration, but can not detect any variation in these conditions. The high sensitivity of the semi-physical models according to the intake manifold composition shows that these models can not be used directly for all the engine control applications.