English a/an appears to be a textbook case of phonologically optimizing allomorphy: it conspires to yield CV syllables instead of hiatus (% ə apple) or extra codas (*an book). But does this effect need to be explained in the synchronic grammar—e.g., is an selected before vowels in order to provide an onset? I argue that it cannot be, based on the selection of an before emphatic glottal stop (an [ʔ]ápple). I provide a serialist analysis of a/an in which allomorphy strictly precedes phonology and cannot ‘see’ surface phonetic forms.