The suspension system of a car is of vital importance for the safety of its occupants. Therefore, it is very important to develop reliable tests for inspecting the condition of its components. A simple model to identify the parameters of a car suspension system is proposed in this paper. It is proven that these parameters are identifiable by using only non-intrusive signals. Unfortunately, the application of conventional identification methods produces suspension parameters without physical meaning. The reason is the loss of consistency of the estimators due to the presence of unknown noise and unmodeled dynamics. In order avoid this effect, the distance between the magnitude of the true and the predicted power spectrum density of the output signal is chosen as the objective to be minimized on a bounded search space with physical meaning. The optimization problem is solved using the MRCD genetic algorithm. Promising results have been obtained for several real-world cases.