The aim of this diploma thesis is to find a solution to evaluate the performance of complex systems such as federated identity management (FIdM) system. We focused on FIdM system, because with the growth of Internet, they would probably be eventually used in order to reduce the number of digital identities (Web accounts) a user needs to have to access web applications. However, no performance evaluation about this kind of systems has been done yet, that´s why, since it is important to have an idea about it in order to implement it correctly, we propose here a way to do it. We concentrated our work on the modelling methodology, because it is a solution whose cost, in term of room and money, is very low, but above all, because modifying a modell in comparison with a real system is very easy. This is important when analyzing the behavior of a system for different sets of parameters for instance. However, before using a model, its accuracy should be tested, i.e. it should be investigated if it behaves like the system it represents. This is the scope of this work. We first looked for the best solutions to model FIdM systems and found two potential models: Queueing Petri Net (QPN) and Layered Queueing Network (LQN). We used them to model an implemented FIdM system and compared their behaviors in order to estimate the validity of the models. The results of the simulations with the two models and the results of the measures made on the real system have showed that both models behave like the real system, as long as they are not overloaded. In surcharged situation, they respond differently: E.g. while the real system drops a lot of requests, if it is overloaded, the models processes them all. Besides, the models are strongly influenced by the slowest component of the system, what is less obvious in the reality. These observations hightlights the fact that the model considered at this step of the work were too simplist to model accurately the reality. As a result, the final conclusion of this work is that the modelling of FIdM with QPN and LQN gives good results, but as soon as the system is overloaded, they must be considered carefully. As already mentioned, the models considered in the work are too simplist and should be improved in the future.