Abstract: Multi-view outlier detection uses the complementary information of each view to obtain more valuable outlier information than single-view. Traditional methods use view connection technology to convert multiple views into single-view, which ignores the complementary information between multiple views. To solve this problem, a multi-view outlier detection algorithm based on total entropy is proposed. The multi-view fuzzy clustering method was used to learn the independent membership matrix and the unified membership matrix of different views respectively to obtain complementary information while preserving the differences between views. The membership matrix was expressed as a binary matrix, and the holoentropy difference was introduced to measure anomaly score for each instance in the cluster. The proposed method was compared with several existing methods on the UCI dataset. The result show that the proposed method has better accuracy and stability.