We propose a method to summarize threaded texts, particularly online discussions (forums and blogs) and e-mail conversations. These texts have semantic relations between their sentences since each post replies to another one. While cross-document structural theory (CST) concerns about links in documents, this work examines the use of CST in summarizing threaded document since no one use it in this domain before. To do so, this work has two phases: first, we identify and extract four CST relations by using supervised machine learning. Second, we develop a new sentence weighting method based on model selection technique over the identified cross-document relations. The experiments show that using of CST in the thread summarization gives promised results. The performances of all methods were evaluated using ROUGE—a standard evaluation metric used in text summarization.