Pollutant emissions and fuel consumption of passenger cars are usually assessed as a function of average as a function of average speed. Nevertheless a number of attempts were made to take into account the driving pattern using emission vs. instantaneous speed and acceleration models. Five quite similar models were developed in Europe. To which extent these instantaneous models are innovative as compared to conventional models? In a first step the various development stages of an emission-based model and the various associated errors are presented. Among the possible sources of error, we selected modelling-induced errors, enabling to compare rigorously the various model types in terms of performance. The analysis was performed using the Modem model, developed from the measurement results obtained over a sample of 150 vehicles, representative of French, English and German fleets, over 14 representative driving cycles under urban conditions. Average emissions as measured over these 14 cycles are compared to emissions calculated over same cycles using this model. Modelling errors range from -51% to +57% as a function of the considered cycle and the vehicle type. This inaccuracy is quite similar for the other four European models. This demonstrates the low reliability of the models used, which cannot be used to assess the impact of slight changes in the driving pattern, sometimes leading to completely false conclusions. These models are barely more precise than average speed based models. In a second step, a number of alternatives liable to improve the reliability of instantaneous models are contemplated: for example using another method for calculating acceleration, or increasing the number of speed and acceleration classes significantly. But this does not improve significantly the model reliability. Limited measurements performed on a catalyst vehicle demonstrated that very high engine loads, even if not frequent, play a significant role in emissions: sometimes they yield emission values a thousand times higher. It is therefore of prime interest to study them in a comprehensive manner. In addition, this demonstrates that each model should be developed from measurements carried out over a set of representative driving cycles under real-world driving conditions.