Abstract：Firstly, a strict mathematical definition of co-occurrence similarity between categorical attribute values was given. Secondly, three other equivalent definitions were proposed. Then, the definition of the co-occurrence similarity between attribute values was extended to calculate the co-occurrence similarity for data objects, and was applied in clustering ensemble successfully. Using the co-occurrence similarity between data objects, the individual similarity matrix of an initial clustering result can be calculated by taking other initial clustering results into account. The experimental results show that Co-occurrence Similarity based on Clustering Ensemble (CSCE) method can effectively identify the subtle structures in data, and improve the accuracy of clustering ensemble greatly.