The study of the language used in Web 2.0 applications such as social networks, blogging platforms or on-line chats is a very interesting topic and can be used to test linguistic or social theories. However the existence of language deviations such as typos, emoticons, abuse of acronyms and domain-specific slang makes any linguistic analysis challenging. The characterization of this informal writing can be used to test the performance of Natural Language Processing tools when analysing Web 2.0 texts, where informality can play an important role. By being one of the most popular social media websites, Facebook handles an increasing volume of text, video and image data within its user profiles. In this paper, we aim to perform a qualitative analysis of informality levels in textual information publicly available on Facebook. In particular, this study focus on developing informality dimensions, a set of meaningful and comparable variables, discovered by mapping textual features by affinity and using unsupervised machine learning techniques. In addition, we explore the relation of informality and Facebook metadata such as received likes, gender, time range and publication type.