Abstract: Association rules are traditionally defined as of the form AB. This form is extended to other three forms AB, AB and AB. A simple but efficient method is proposed to calculate the support and confidence of the three forms. Some problems such as self-contradictory rules may occur when studying both the positive and negative association rules simultaneously. These problems are discussed and the corresponding solution by correlation is proposed. An algorithm is also proposed to mine both positive and negative association rules from frequent itemsets. An experiment is performed and the experimental results demonstrate that the algorithm can detect and then delete those self-contradictory rules.