The invention discloses a method for automatically segmenting human lateral geniculate nucleus (LGN) by using prior knowledge. The method comprises the steps: firstly, carrying out bias field correction on data of a structure image of a human brain, and then, carrying out brain substructure segmentation on the corrected image, so as to obtain a template of a LGN; then, registering the corrected structure image data and the template of the LGN to a MNI (Montreal Neurological Institute) standard space; acquiring the region boundary of the LGN in the MNI space according to prior knowledge of an LGN anatomical structure, and then, segmenting the LGN in the region respectively by a region-growing method, a k-means method, an Otsu algorithm and an image segmentation method; then, fusing the segmented results, so as to obtain an estimated value of the LGN region, which serves as a segmented result; finally, transforming the segmented result into an original space of the structure image, thereby obtaining a final LGN segmented result.