Under the framework of visual grammar developed by Kress, Van Leeuwen, Painter, etc., this paper conducts a multimodal discourse analysis on the Photos section of US Army official website. The analysis is conducted from three perspectives of representation, interaction and composition meanings, with the purpose of investigating the semiotic resources of the webpages, exposing the mechanism for multimodal discourses togenerate meanings, exchange information and construct images. Informed by the genre theories established by Halliday, Hasan, Bateman, Bhatia, etc., this paper also investigates the inter-mode relationships, the overall layout, generic features and publicity strategies of the website. It is found that the website intends to manipulate readers’ perception of the US military, construct its positive images and neglect negative issues through using symbolic meanings, emotional interaction and salience difference. After an in-depth analysis of the communicative purpose, the generic structure potential and the interaction mechanism, a multidimensional framework to analyze the multimodal genre is proposed to reveal the complex relationship between the design of military websites and the social cultural contexts in order to help military image construction.