This study reports on the results of classroom research investigating the effects of using data-driven learning methods by students in revising their writing errors. The main purpose of the study is to examine to what extent is consulting a corpus effective in correcting lexical errors in their writing. It has been found in previous research that data-driven learning can benefit students in the revision process and that it works better for certain error types (Tono et al, 2014). The targeted error category was lexical errors including formal errors or semantic errors. For the study, a small corpus of 44 student paragraphs written in a timed writing task was used. The corpus of student paragraphs was analyzed for the common lexical errors. The error classification used in the study was drawn from the lexical error taxonomy of James (1998). All lexical errors were hand tagged according to the taxonomy. From among the common errors, certain errors were chosen for revision activities. Students were given hand-on instruction on using an online corpus and its concordancing tools and were asked to revise the selected errors by referring to the corpus. The effectiveness of consulting a corpus while revising errors was compared for different lexical error types.