To solve the problem that architects were always having problems in finding suitable architectural images effectively from websites, a semantic acquisition method of architectural images based on web annotation technology was proposed. First, the concepts and types of architectural images and semantics were defined. Second, the framework and operational processes of this method were illustrated. Finally, the feasibility and validity of this method were verified by famous building websites as examples. Regarding to operational process, there were three steps: building the architectural semantic dictionary by adding artificially and learning online; collecting 6 items of image-related texts (image's name, image's annotation, information around images, webpage's title, webpage's body, and the title of image's hyperlink) from websites where the images were founded; collecting semantics of images according to certain rules, relating them to image files and keeping them to the database of architectural images. This study indicates that the proposed method is feasible and easy to be operated. Architectural images can be automatically downloaded in batches and more than 30 items of architectural semantic characteristics will be collected, such as names, categories, themes of architectural images, as well as names and types of projects. As a result, the semantic acquisition method of architectural images will effectively overcome the problem that architectural images are hard to be searched and it will help architects to improve their abilities of innovation by using images from websites.