In my presentation I will describe a supervised machine learning approach for summarizing legal documents. A commercial system for the analysis and summarization of legal documents provided us with a corpus of almost 4,000 text and extract pairs for our machine learning experiments. We pre-processed that corpus to identify the selected source sentences in extracts from which we generated legal structured data. We then ran a sentence classification experiment relying on a Naive Bayes classifier using a set of surface, emphasis, and content features. Finally we compared our results against a baseline and the commercial system being used currently. [Ce séminaire est une "avant-première" d'une présentation à faire à Canadian AI 2010]