A method and an apparatus for recognizing a 3D image. The method comprises the procedures: a: inputting an image; b: processing the image into a left gray level image and a right gray level image; c: performing, by using a corner detection algorithm, processing on the left gray level image and the right gray level image to separately solve corners; d: matching corners of the two images by using a corner matching algorithm; e: eliminating a wrong matched corner by using an elimination algorithm; and f, determining whether the number of pairs of matched corners of the two images is greater than or equal to N; if yes, the image being a 3D image; and if not, the image being a non-3D image. By means of the foregoing mode, the provided method and apparatus for recognizing a 3D image can automatically distinguish a 3D image and a non-3D image, without the need for a user to carry out eye discrimination, thereby saving the time and energy of the user.