The invention provides a hyperspectral image anomaly detection method using multi-window feature analysis. The hyperspectral image anomaly detection method comprises the following steps: at first, determining the size of detected windows including an inner-layer window, a middle-layer window and an outer-layer window; next, calculating an OSP (Orthogonal Subspace Projection) operator in the outer-layer window, eliminating background interferences in the inner-layer window and the middle-layer window, and effectively removing white noise; then, carrying out background image element selection in the middle-layer window; and then, calculating a KRX (Kernel RX) operator in the inner-layer window, and carrying out anomaly detection on an image element to be detected; finally, outputting a detection result. According to the hyperspectral image anomaly detection method, a detection mode for three layers of windows is skillfully applied, and hyperspectral data is subjected to noise interference elimination at first and then is subjected to anomaly detection by using two layers of local background pixel windows. The interferences or the white noises emitted by uninterested signal sources in the inner-layer window and the middle-layer window are eliminated by using the OSP operator in the outer-layer window, so that the false alarm probability is reduced and better detection effect is obtained. A simulation experiment is carried out by using AVIRIS (Airborne Visible / Infrared Imaging Spectrometer) hyperspectral data, the detection performance of the hyperspectral image anomaly detection method provided by the invention is remarkably superior to the traditional algorithm, the false alarm possibility is reduced, and better detection effect is gained.