To date, few studies have attempted to formulate typologies of errors by non-native speakers (NNS) in English scientific writing. In this study of 123 doctoral dissertation abstracts written by doctoral students in France, we present a tentative typology of frequent errors that covers issues with general grammar, expert grammar and style. In order to specifically ascertain the errors made by students who experience very significant difficulties, the 123 items of our corpus were chosen after an initial review of 1018 abstracts because they demonstrated low linguistic and stylistic proficiency. The typology of errors was sought in support of an error identification exercise in the Scientific Writing Assessment Program (SWAP), an English language certification recently developed at ENS Paris-Saclay.1 Although some disciplinary variation was seen in the distribution of errors, a convergence towards six major error types (determiners, syntax, tense choice, compound phrases, collocations and lack of clarity) was observed (62.96 % of all errors in geoscience, and 83.89 % in mechanical engineering), suggesting that efforts to mitigate errors should primarily focus on these key issues. Another key finding was that, in contrast with previous studies, traditional grammar issues did not represent the bulk of overall errors (52.78 % in geoscience and only 37.32 % in mechanical engineering), while the overall frequency of stylistic errors was high in both corpora (30.25 % in geoscience, 46.05 % in mechanical engineering), showing the importance of errors in relation to genre-specific style. We propose a metric of error frequency, the Comprehensive Error Ratio or CER, to assess the overall quality of abstracts written by non-native speakers of English. In conclusion, we suggest that any typology of errors in ESP/EAP contexts results from a trade-off between seeking descriptive specificity and achieving the specific purposes for which a typology is developed.