Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for web page identification by propagating scores in a graph. In one aspect, a method identifies two or more web pages, wherein each web page is associated with one or more features, and each web page is associated with a respective first score, and each feature is associated with a respective second score. A graph is generated linking the two or more web pages and the one or more features. An updated first score is calculated for each web page and an updated second score is calculated for each feature. For each web page, a determination is made as to whether the respective updated first score meets a threshold. The web page associated with the updated first score that meets the threshold is identified as a web page that an advertiser can use as a landing page.