This project, funded by the Nuffield Foundation, examines the lexical features of children's books using corpus analysis. We created a picture book corpus from 160 texts commonly read to children aged 0-5 years. We first quantified how the language of these books differs from child-directed speech (compiled from 10 corpora in the CHILDES UK database) on measures of lexical richness (diversity, density, sophistication), part of speech distributions, and structural properties. We also identified the words occurring in children’s books that are most uniquely representative of book language. We found that children’s book language is lexically denser, more lexically diverse, and comprises a larger proportion of rarer word types compared to child-directed speech. Nouns and adjectives are more common in book language whereas pronouns are more common in child-directed speech. Book words are more structurally complex in relation to both number of phonemes and morphological structure. They are also later acquired, more abstract, and more emotionally arousing than the words more common in child-directed speech. Written language provides unique linguistic input even in the pre-school years, well before children can read for themselves.