This paper investigates the (re-)emergence of onset consonants in English loans from French, Latin and Greek, spelt with initial ‹u› (> /juː/; e.g. union, use), initial ‹eu› (> /juː/; e.g. eulogy, euphemism), or initial ‹h› (e.g. habit, homogeneous). It analyses Google Books data, exploiting the occurrence of the article allomorph a (rather than an) as a diagnostic of consonantal realisation. The analysis yields a fine-grained description of the (re-)emergence of consonantal onsets. It shows that their emergence has been a gradual process and has not reached completion yet. On a theoretical level, the paper discusses the interaction between categorical phonological processing and fine-grained phonetic distinctions in an exemplar-based framework. It also sheds light on the question of (near-)mergers and their potential reversibility.