With corpora firmly established as a research tool in the field of linguistics,their use for pedagogical purposes has also been increasingly advocated.Recent years English writing instruction has witnessed a growing interest in the use of corpora in China.Many researchers have explored the potential of incorporating corpora into English writing instruction,which ranges from presenting learners the printed concordance lines to guiding students to discover language features by carrying out tasks by corpus themselves.However,aspects such as students' attitudes and perceptions of the effectiveness corpora in writing instruction and difficulties and barriers experienced by teachers and students in using corpora remain less studied in the context of China.This paper described a small-scale study of corpora use in the EFL writing class of tertiary level.Specifically the study examined the students' and teachers' attitudes and perceptions of using corpora to facilitate the acquisition of English collocation in writing.The participants of the study are three teachers and 90 first and second year students involved in a 16-week selective English writing course in a university in China.At the beginning of the course,students will receive training on how to use corpora for selfdiscovery learning.After the training,students are asked to find out their own difficulty in using particular collocations for each writing assignment.Once identifying their own problems,students are required to solve their problems through the use of corpora.At the last stage,students are also required to summarize the usage of the collocations as a part of their assignments.The study employs a variety of data collection methods which include questionnaires,interviews,students' and teachers' reflection journals.A close analysis of the data reveals the students perceived effectiveness of the new approach as well as teachers' and students' attitudes towards the use of corpora.Furthermore,difficulties and challenges experienced by both teachers and students are also revealed from the data analysis.