The analysis of social media is currently very important due to the unprecedented quantity of information. Twitter is becoming an indispensable source of information for researchers aiming to implement big data in their projects. However, despite the potential eld of research opened by that Twitter data, it contains some risks a researcher must be aware. In this paper I present on the one hand the bene ts and caveats of research conducted on Twitter, and on the other hand the constraints of Twitter data collected from the Application Programming Interfaces (APIs). There are, therefore, three major methodological problems identi ed: (i) representation bias: it is very di cult to make general assumptions using research based on Twitter. (ii) language challenge: users can write in many di erent languages. It implies that when collecting data, some cautions need to be taken in order to accurately gather the data we need, (iii) data bias: Depending of the data needed, one API might be a better t than other. The main aim in this paper is to discuss these methodological constraints from a theoretical point of view. I propose, as a starting point, possible solutions to overcome them, or at least reduce their impact in the research.