Outlier detection is an important content in data mining and is widely used in the field of credit card fraud detection, network invasion detection and so on. According to hierarchical clustering and similarity, this paper presents the concept of high dimensional data similarity measurement function and class density, based on class density, the outlier of high dimensional data is redefined so that a kind of outlier detection algorithm based on similarity measurement is proposed. Experiment shows that this algorithm has certain value on outlier detection in high dimensional data.