This paper studies the role of images in online news, showing how news media exploit leading pictures to influence readers’ processing of the issues. I document two relevant facts. First, the US news’ visual language is distinctive of the sources’ political leaning and significantly polarized, to an extent comparable to the documented verbal polarization of US Congress in recent years. For this analysis, I construct a visual vocabulary of graphic features and apply a dictionary-based method to study the visual language polarization in the leading images published in US news between December 2019 and December 2020. Second, such visual partisanship is an expression of political media bias: in a survey experiment, individuals exposed to identical news pieces but leading pictures with opposite partisanship formulate significantly different opinions, which are slanted towards the images' respective ideological poles. I find that news’ visual bias causes a significant increase in issue polarization of the public. The slanting effect of images interacts with readers’ prior, and audiences on both sides of the political spectrum react more distinctly to pictures aligned with their viewpoint. This pattern implies that the polarizing effect of visual bias is further exacerbated if readers' source their news exclusively from like-minded outlets, as in information echo-chambers.
[Draft available upon request]