The need for traffic noise prediction models that take traffic dynamics into account has been recently shown for urban areas. Such models couple a dynamic representation of traffic with noise emission laws. The contribution of the paper is to test different traffic and noise source representations for LAeq and statistical levels estimation. Tests on four scenarios that reflect urban traffic conditions are carried on. They show that an individualized representation of vehicles with a macroscopic behavior rule is sufficient for noise descriptors estimation. Noise sources have to be aggregated on cells to reduce the calculation time of noise emission propagation. To this end a grid of line source representation appears to be more relevant than a grid of point source representation. Furthermore, large cells do not affect substantially the noise descriptors estimation.