Reflecting a long history of contact between the Japanese and English languages, a large number of English loanwords have become integrated into the general, everyday Japanese language. This study is a corpus analysis of the grammatical behaviour of frequently-used English loanwords in contemporary Japanese. It addresses a previous lack of research in the area by providing the first, large-scale, empirically-grounded account of such grammatical behaviour. Framed within a lexico-grammatical view of language, a sample of over 500 English loanwords were analysed within their naturally-occurring linguistic contexts in a large Japanese corpus. For this, corpus analysis software was used to generate a ‘word sketch’ for each loanword showing their most frequent grammatical relationships and their most salient collocates in each relationship. The word sketches were collated into a database of over 5000 grammatical relationships and then compared to a database of over 1000 grammatical relationships of native and Sino-Japanese words. The comparison revealed a marked pattern of behaviour of the loanwords, with a large number strongly favouring a compound noun grammatical relationship. A subsequent analysis of the most salient collocates of a sub-sample of the loanwords found that the more strongly a loanword favoured the compound noun grammatical relationship, the more strongly and exclusively it collocated with other loanwords rather than with native and/or Sino-Japanese words. In accounting for this behaviour, these loanwords appear to be ‘non-catachrestic innovations’ (Onysko and Winter-Froemel, 2011), a category of loanwords which are seen to be the pragmatically marked lexical choices in a language. With these findings, this study contributes to a more thorough empirically-grounded understanding of the interaction between the Japanese and English languages, as well as to a reconceptualisation of the grammatical integration of loanwords in a language.