In a crossmodal priming experiment, visual targets (e.g. RENARD, 'fox') were auditorily primed by either an intact [see text] 'the fox' or reduced form [see text] 'the fox' of the word. When schwa deletion gave rise to an initial cluster that respected the phonotactic constraints of French (e.g. [lapluz] 'the lawn' in which /pl/ is a legal word beginning in French), there was a processing cost for the targets primed by the reduced form of the word compared to intact primes (e.g. [see text] 'the lawn'). However, when schwa deletion produced an initial cluster that violated the phonotactic constraints of French (e.g. [see text], where /Rn/ is not allowed as a word beginning), there was no penalty for targets primed by reduced compared to intact forms of the word. Assuming that listeners change their phonemic percept when confronted with phonotactically illegal sequences (Segui, Frauenfelder, & Hallé, 2001), phonotactic constraints may help to restore the deleted schwa in sequences like le renard [see text] in French.