Genetic algorithm-based learning classifier system (LCS) is a massively parallel, message-passing and rule-based machine learning system. But its potential self-adaptive learning capability has not been paid enough attention in reservoir operation research. In this paper, an operating rule classification system based on LCS , which learns through credit assignment (the bucket brigade algorithm) and rule discovery(the genetic algorithm), is established to extract water-supply reservoir operating rules. The proposed system acquires the online identification rate 95% for training samples and offline rate 85% for testing samples in a case study, and further discussions are made about the impacts on the performances or behaviors of the rule classification system from three aspects of obtained rules, training or testing samples and the comparisons between the rule classification system and the artificial neural network (ANN). The results indicate the learning classifier system is feasible and effective for the system to obtain the reservoir supply operating rules.