For scientists in the twenty-first century, a mastery of written English for Research Publication Purposes (ERPP) is fundamental to professional success. For many Japanese scientists, this is a source of frustration given their use of English as a foreign language is subject to idiosyncrasies, which may be perceived by editors, reviewers, and readers as errors detracting from the impact of their research. As a basis for building data-driven needs-specific pedagogical support, this study investigates the major error patterns in Japanese scientists' written English. Participants in the study are 13 Japanese scientists working in the field of materials science. The primary data are the participants' scientific research article manuscripts (i.e., the research articles before publication). An elaborated corpus-assisted Error Analysis (EA) methodology is employed, investigating error patterns through the lens of Systemic Functional Linguistics (SFL). In short, the EA involves the collection of texts, identification of errors in the texts, and the classification, linguistic description, and quantification of those errors, with these classifications and descriptions elaborated through the SFL theory. For purposes of scope, the investigation focuses on errors in nominal groups—a key feature of scientific English. A total of 654 nominal group errors are identified and analysed according to the elaborated EA methodology. From the analyses, two major error patterns emerge: errors with articles and plural -s, and errors with preposition -of. Results highlight the difficulties Japanese scientists face with both the pre- and post-modification of complex nominal groups. Particularly, the omission of the indefinite article a with singular referents, and the misselection of embedded -of prepositional phrases in post-modification are revealed as dominant errors in this study. Detailed descriptions of the two major errors patterns and their implications along with recommendations for pedagogical intervention are presented.