Most of the existing hyperspectral anomaly detection algorithms only use the spectral dimension information of hyperspectral images, which does not reflect the advantages of “image-spectrum integration” of the hyperspectral data, and may result in degraded detection performance.This paper proposes a hyperspectral anomaly detection algorithm based on non-local self-similarity（NLSSAD）.The algorithm first establishes dual stereoscopic windows, in which the inner window represents the 3D spatial-spectral structure window of the Spectral Vector of the Pixel to be Tested (SVPT).Then the stereoscopic window in the background is found, which is the most similar to the inner window, and the distance between the two windows is calculated to obtain the non-local self-similarity index of SVPT.The anomaly detection results show that, compared with the existing algorithms, NLSSAD has better performance in detection rate and operation speed.