||Since several years, a considerable number of researchers has been concerned with the automatic optimization of the parameters of fuzzy systems. While almost all of them considered offline methods, leading to optimization problems in the usual sense, only a very small minority focused on online methods, where the rules (possibly even the fuzzy sets) are modified according to responses from a given (simulated or real) environment in a closed-loop architecture. The main branch of such methods is based on so-called Classifier Systems (CS).The aim of this thesis is to examine the functionality of a classifier system combined with a fuzzy system - the so-called Fuzzy Classifier System (FCS) - a machine learning procedure that tries to unify the advantages of fuzzy systems with the robustness of genetic algorithms. For this examination, an object-oriented framework has been designed which is capable to simulate various kinds of classifier systems. The chosen language for this implementation was C++. The framework structure is kept in a way to allow easy extension, so it is capable to deal with future renewals in this area.The Fuzzy Classifier System is a learning method based on reinforcement learning. To understand the performance of the FCS, an introduction to genetic algorithms is given, which is an important component of this method. Also the necessary foundation of fuzzy systems will be presented. Then the thesis goes into detail and describes the structure and the performance of the fuzzy classifier system. Some alternatives to the FCS are discussed, but are only touched superficially. Some problems and disadvantages of the FCS are discussed, mainly the selection scheme and the overall results of the FCS. A possible solution to these problems is presented - a modified version of the FCS. After these explanations, some results of the tested framework are shown and discussed.