Abstract: An efficient algorithm, RIFCI, for mining association rule, is proposed by simultaneously mining the frequent closed item sets instead of traditional frequent item sets and at the mean time exploring both the item set space and transaction space, as is distinguished from all previous association mining methods that exploit only the item set search space. Compared with the industrial standard C4\^5, the error rate of the result is less than 19\^48%, got by searching ten data sets in UCI repository of machine learning database. This algorithm, without changing the accuracy, generates the number of roles lower than previous methods, which improves the efficiency.