Outlier detection in high dimensional data is one of the hot areas of data mining. The existing outlier detection methods are based on the distance in Euclidean space. In high-dimensional data, these methods are bound to deteriorate due to the notorious "dimension disaster" which leads to distance measure cannot express the original physical meaning and the low computational efficiency. This paper improves the method of angle-based outlier factor and proposes the method of variance of angle-based outlier factor outlier in mining high dimensional. It introduces the related theories to guarantee the reliability of the method. The empirical experiments on synthetic data sets show the method is efficiency and scalable to high-dimensional data sets.