In most academic writing programs worldwide, instruction aims to develop students' awareness of essay types, written language use, and test-taking strategies. However, teachers' instructional practice is often dictated by their knowledge base, perception, and the intuitions they have about language use (Borg, 2006). Such intuitions are not always correct (Cobb & Boulton, 2015) and may not fully reflect how language is used in a particular type of academic writing by native speakers. Thus, the grammatical structures they teach may not correspond well to what is frequently and naturally used in that type of text. On the part of students, because of their limited knowledge of writing conventions, they might end up using a variety of grammatical forms while not considering how appropriate these are. The authors feel that it is imperative for writing teachers to allow for 'data-driven learning' (Johns, 1991, p. 17) whereby students explore language constructions through consulting a purpose-built corpus. Such a corpus exposes students to language features peculiar to a particular written genre, but it can also help them identify which are more frequently and appropriately used than others in the genre. In this article, the authors describe how to identify frequently used complex sentence forms in a corpus of essays written by experts to inform the design and implementation of teaching materials for writing.