Due to substantial commercial benefits in the discovered frequent patterns from large databases, frequent itemsets mining has become one of the most meaningful studies in data mining. However, it also increases the risk of disclosing some sensitive patterns through the data mining process. In this paper, a multi-objective integer programming, considering both data accuracy and information loss, is proposed to solve the problem for hiding sensitive frequent itemsets. Further, we solve this optimization model by a two-phased procedure, where in the first procedure the sanitized transactions can be pinpointed and in the second procedure the sanitized items can be pinpointed. Finally, we conduct some extensive tests on publicly available real data. These experiments’ results illustrate that our approach is very effective.