The use of discourse markers in argumentation is an area in which the rhetorical traditions of Arabic and English diverge. While English allows implicit logical links, Arabic demands explicitly marked coherence. Such differences in the expression of discourse relations are brought to the fore when these two languages come into contact through translation. This paper reports on an empirical investigation into the translation performance of English–Arabic translation students at different stages of a postgraduate translation programme. A preliminary quantitative exploration of the translation product is combined with a qualitative analysis, in which translation process data is interpreted in light of Relevance Theory. Think-aloud protocol transcripts reveal that advanced translation students supplement their intuitive relevance-based comprehension mechanism with a conscious, analytical processing ability of a more deliberate nature. The results both demonstrate the problematic nature of this aspect of translation and highlight the need for explicit classroom instruction.