Causal correlation method was one of the most representative methods for instruction detection alert correla-tion. In some conditions

the correlation graph would be split because of loss of causal information. In order to solve the problem

an algorithm was proposed to reconstruct attack scenario using fuzzy clustering. A new similarity membership function based on the attribute hierarchy tree was defined in the process of clustering. Furthermore

the evaluation method and indexes were put forward to describe the ability of reconstructing attack scenario. The experimental results indicate that this algorithm is valid to combine the split correlation graph and reconstruct attack scenario.