Predicting the eventual judgement in legal cases requires consideration of more than just the facts since judges also consider the quality of the argumentation presented by the parties. Most prior NLP work focuses on just the facts and overlooks the essential element of legal argumentation within the court process. Working toward argument extraction and comparison technology, this paper describes the construction of a dataset comprising 8364 cases, including judgements. The arguments supporting each allegation have been identified and preprocessed. We show that there is no trivial solution to legal judgment prediction using simple correlation features. Facilitating a novel machine learning perspective on the Legal Judgment Prediction task, this is the first dataset of its kind; previous datasets do not provide this level of argumentation detail.