Abstract : -Classifier systems are rule-based control systems for the learning of more or less complex tasks. They evolve in an autonomous way through solution without any ex-ternal help. The knowledge base (the population) con-sists of rule sets (the individuals) randomly generated. The population evolves due to the use of a genetic algorithm. Solving complex problems with classifier systems involves problems to be split into simple ones. These simple prob-lems need to evolve through the main complex problem, 'co-evolving' as agents in a multi-agent system. Two different conceptual approaches are used here. First is Elitism that is inspired by Darwin, distinct agents evolving always keeping alive their best members. Second is Dis-tributed Elitism which is a logical enhancement of Elitism where agents knowledge is distributed to make the whole evolve through solution. The two concepts have shown in-teresting experimental results but are still very different in use. Mixing them seems to be a fairly good solution.