In this paper artificial immune recognition system (AIRS) is employed as an emerging technique of data mining to extract the reservoir operating rules with a case of water supply reservoir, and we mainly focus on the impacts of learning mechanisms of AIRS on the obtained operation rules, therefore the mechanisms are explored and different gene encodings, as knowledge representatives, and the uncertainties of annual hydrological conditions (AHC), one attribute of the operating data, are considered. In order to further illuminate the learning capabilities, the classification results of the rules through AIRS and RBF networks are compared, indicating AIRS can be better for mining the reservoir operating rules which are of more transparent and interpretive, and can be dynamically updated.