In order to reduce storage losses, extend the storage period and improve the storage efficiency of winter jujubes, taking the winter jujubes in Zhanhua City as research object, a hyperspectral imaging system was built for detecting bruises happened less than 1h on ‘Zhanhua’ winter jujubes. The 256 hyperspectral images data within 871~1766nm wavelengths were obtained by the hyperspectral imaging system. The effective wavelengths were selected by using uninformative variables elimination (UVE) method and correlation coefficient method. Eventually, four characteristic wavelengths, 944, 1035, 1187 and 1376nm were selected. Four images were mapped to four characteristic wavelengths, the principal component analysis (PCA) was used based on the four images, and the first component of the image was selected to future process and segment. Many pretreatment methods were used for the first component of the image, such as gray level transformation and so on, and the method of adaptive threshold was applied to segment. Finally, the slightly damaged area could be obtained. Thus, the slight bruises on winter jujubes were recognized. Independent validation set of 100 bruised winter jujubes was used to evaluate the performance of the developed algorithm. Results showed that 98% of the bruised winter jujubes were recognized correctly.