Today's advanced driver assistance systems (ADAS) are increasingly becoming more complex. The next step in this direction is the development of automated driving systems. However, along with the complexity, the effort required for development and validation of these systems is increasing as well. In order to be able to master this complexity in terms of cost and time, simulations are being used more frequently. Since perception sensors are essential for highly and fully automated systems, it is important to have valid sensor models for the development and validation of such systems. Therefore, in this paper we present a method for the validation of perception sensor models. The method differs from known approaches by its two-step procedure. First we directly compare the experimental results and the model output. Then the output of the next tier in the system, using real and synthetic data from sensor model as input, is compared. We also show the tool chain for implementation along with first results, confirming the potential of our approach.