Visual analysis represents a significant oversight in the corpus literature, and possibly one that may lead to unintended omissions, particularly when analysing social media. In this chapter we introduce Visual Constituent Analysis (VCA), a method of multimodal corpus construction that allows researchers to construct and analyse visual aspects of online media in large-scale corpora. The chapter addresses the shortcomings of a purely textual approach to discourse analysis when dealing with social media texts and offers a solution using computer ‘Vision’-based image annotation (in our case Google Cloud Vision). Finally, we demonstrate how our approach can be used to analyse a sample of 150,000 micro-blog posts from Twitter and show the difference in level of user interaction with combined image/texts over language-only social media texts.