Due to the increasing number of electrical and electronic system in cars and the fact that all of them have to operate with limited electromagnetic interference, the automotive industry has to deal with an emerging problem: how to simulate as accurately as possible electromagnetic fields in a car body during the design phase while minimizing computation time and costs?EM simulation of components or systems over a wide frequency range raises a variety of problems set aside its inherent complexity. Not only the simulation must take the frequency domain into account it has also to deal with the spatial discretization step. The parameters that the EM simulations are relying on have to be automatically adjusted to the complexity of the CAD data.Therefore to be able to realize an automated translation of CAD data to EM simulation data we propose a new methodology for CAD part approximation and gathering. The input parameters are the fundamental frequency of the simulation to be carried on, the model overall size and the level of detail to be passed to the approximation process. In order to pursue the approximation of data (surface or volumes) the process has to deal with CAD data quality (geometric and/or topologic criterions), neighborhood relationships and parts assembly tree.We will demonstrate in concrete terms through a couple of examples how the process greatly improved the overall time required to carry the simulation through.