Apriori algorithm is a classical association rule mining algorithm, but it has problems about frequently scanning database and generating a large number of candidate sets. To solve these problems, an improved DC_Apriori algorithm was proposed, which restructured the storage structure of the database, improved connection of frequent item sets, the generation of k-frequent item sets is only need to join the 1-frequent item sets with k-1-frequent item sets, greatly reduced the number of connections and it can directly get frequent item sets by only one pruning operation, effectively avoid the unnecessary invalid candidate sets, and greatly reduce the number of scanning the database and improve the efficiency of frequent item sets generation. It has proved by experiments that the DC_Apriori algorithm is obviously superior to the Apriori algorithm and the MC_Apriori algorithm based on the matrix, whether in small support degree or in the intensive database with large numbers of transactions and items, the running time of DC_Apriori to get the same result is significantly less than the Apriori algorithm and MC_Apriori algorithm based on the matrix.