Corpora have been used in English as a foreign language materials for decades, and native corpora have been present in the classroom by means of direct approaches such as Data-Driven Learning (Johns, T., and P. King 1991. "'Should you be Persuaded'- Two Samples of Data-Driven Learning Materials." In "Classroom Concordancing," 1-16. Birmingham University. English Language Research Journal 4.). However, the suitability of using learners' output in classroom tasks remains controversial. This paper describes a pilot study in the application of a non-standard direct approach where Spanish university students are invited to reflect on their production. In the experiment, carried out in several sessions during the course, the students were exposed to a selection of erroneous sentences from their compositions. Prior to the classroom activity, the teacher contrasted the learners' sentences with correct versions produced by native English speakers. A relevant part of the methodology consisted in getting learners collectively involved in finding the errors and suggesting improvements. After that, solutions were discussed through the analysis of the alternative sentences provided by the native students. The results show that students are willing to accept this methodology as a supplement to textbooks' proposals. We claim that authentic and highly specific learner data obtained from a reliable ad hoc learner corpus and direct exposure to these data through controlled activities may cover certain learners' needs not found in textbooks.