A method for studying use of scientific sources in arguments on Twitter is demonstrated. Data were collected from the Twitter API v. 2.0 using Focalevents, searching for tweets with links to DOIs, and then collecting conversations around these tweets. Analysis. Three conversations on different topics were analysed searching for argumentative behaviour, use of scientific sources, their reliability, consistency and adequacy in relation to the argument and the target audience. Both quantitative and qualitative content analysis based on argumentation theory were applied. The method allowed us to identify scientific publications used argumentatively by a multiple audience in the context of Twitter conversations. The publications were used to build scientific arguments, mainly, but not exclusively, from individual and collegial expert opinion. Scientific findings were often misinterpreted and used improperly to the benefit of the argument. Through the use of argumentation theory to study conversations in a structured way, the paper demonstrates how to approach the usage of scientific publications in arguments. Scientific publications were used to build scientific arguments from different types of expert opinion, for giving proofs for claims and ounter-arguments, and inconsistent or biased arguments from individual expert opinion.