Whether the phonological deficit in dyslexia ensues from impaired universal phonological grammar remains unanswered. Ten French dyslexic children were compared to chronological age-matched and reading level-matched controls. All were aurally-administered two syllable-counting tasks with di- and trisyllabic pseudowords (/am.ʒal/; Exp. 1) or mono- and disyllabic pseudowords (/mʒal/; Exp. 2). We manipulated universal phonological sonority-related markedness within unattested onset clusters, from the phonotactically-unmarked clusters (/bʁal/) to the phonotactically-marked ones (/ʁbal/). Sonority-related markedness was “naturally” inverted within intervocalic clusters. Across experiments, as sonority-related markedness decreased, response accuracy increased. Response patterns were similar in dyslexic children who were as accurate as both control groups, but systematically slower. Remarkably, an illusory epenthetic /ə/ vowel phonologically repaired marked clusters. Neither statistical properties nor acoustic-phonetic cues and reading skills accounted for the misperception/repair. Our results discard impaired phonological grammar as a source of dyslexics' phonological deficit and are discussed toward a deviant vs. delayed developmental course.