In this paper we describe a corpus, the Lancaster-Leverhulme Corpus of Children's Writing, which is nearing completion at Lancaster University. The corpus has proved a particularly challenging one to construct, as the age range covered by the corpus, 8–11 years of age, is one in which literacy is still very much developing. As a result we have had to make a wide range of transcription decisions related to such issues as spelling and general corpus encoding. Our response to the data has been to develop a set of minimal encoding principles for the transcription of handwritten data. Additionally, to ensure that as much information as possible from the original is preserved, we are digitally scanning as well as transcribing the children's projects, making this a multimedia corpus.