French (or English) native listeners hear /kl/ when presented with the illegal consonant sequence */tl/. This robust case of perceptual repair is usually viewed as operating at a prelexical level of speech processing but the evidence against lexical feedback is somewhat weak. In this study, we report new data supporting the prelexical hypothesis, obtained with a paradigm that avoids most of the possible confounds in previous studies. In a cross-modal auditory–visual priming paradigm, lexical decisions to the same visual target “clavier” are facilitated by the auditory prime *tlavier, not by *dlavier. Likewise, the recognition of “glacier” is facilitated by *dlacier, not by *tlacier. To summarize, velar stop + /l/ words are exclusively facilitated by the dental-initial derived forms with the same voicing. Derived forms with the opposite voicing tend to induce inhibition rather than facilitation. Hence, the observed facilitation effects are not graded from */tl/ to */dl/ or vice versa. We argue that these rather surprising all-or-none priming effects exclude the possibility that the */tl/→/kl/ and */dl/→/gl/ repairs are due, even partly, to lexical feedback. Index Terms: phonotactic, perceptual repair, lexical feedback