The present study assumes that, despite the relative uniformity of research articles (RAs) imposed by the requirements of the genre, there may be intercultural variation in the rhetorical preferences of different writing cultures. This study develops further Moreno's [Text 18 (1998) 545] model for the comparison of the metatext employed in English and Spanish to signal premise-conclusion intersentential coherence relations. It does so by focusing on the types and preference of use of retrospective cohesive mechanisms employed in premise-conclusion metatext to label the premise from which the upcoming conclusion is to be drawn. Variability is sought in different aspects of the label arriving at the following conclusions: (1) With regards to the extent to which authors make explicit reference to the stretch of discourse from which the upcoming conclusion is to be drawn, Spanish academics show a greater tendency towards the use of fuzzy labels; (2) The overall distribution of the lexical range of labels is also different, with English showing a greater tendency towards the use of non-metalinguistic labels. Retrospective labels have a greater tendency to add interpersonal meanings in English both (3) through the label itself and (4) through its modification; (5) The various ways in which modifiers in retrospective labels add ideational meaning seem to be distributed differently. (Contains 5 tables.)