Data scale becomes a bottleneck in user behaviors analysis, many data mining algorithms become Inefficient slow in this circumstances. This paper explores an effective approach to mine latent knowledge in large scale data, which combines the basal principles of association rules, MapReduce model and Hbase database. First, general principles and algorithm of association rules are given. Second, the work mechanism and traits of MapReduce model and HBase are introduced. Finally, it gives detailed design methods that how to combine the basal principles of association rules, MapReduce model and Hbase database. Sufficient experiments prove that the processing velocity of parallel approach nearly decuple unparallel approach’s. Therefore, the approach combined association-rule and cloud computing is a successful and valuable exploration.