An image retrieval device that retrieves an image in response to an input image by comparing feature amounts based on an attribute of search target images and a feature amount based on an attribute of an input image. Each of the feature amounts comprises a first feature amount expressing a global relative positional relation of each pixel constituting an image with respect to pixels of the entire image, and a second feature amount expressing a local relative positional relation of each pixel constituting each of the images with respect to its adjacent pixels. Component images and at least one combination image formed therefrom are linked to a search target image and feature amounts about these respective images are compared with a feature amount about an input image and a search target image linked to a component/combination image that is similar to the input image is displayed.