An often overlooked but fundamental issue for any comprehensive model of visual-word recognition is the representation of diacritical vowels: Do diacritical and nondiacritical vowels share their abstract letter representations? Recent research suggests that the answer is "yes" in languages where diacritics indicate suprasegmental information (e.g., lexical stress, as in cámara ['ka.ma.[voiced alveolar tap or flap]a] camera; Spanish), but "no" in languages where diacritics indicate segmental information such as a different phoneme (e.g., the German vowels ä /[open-mid front unrounded vowel]/ and a /a/). Here we examined this issue in French, a language that contains a complex set of diacritical vowels (e.g., for the letter e: é, è, ê, and ë). In Experiment 1, using a semantic categorization task, we compared the word identification times to intact diacritical words (e.g., chèvre, goat in English) with a condition with omitted diacritics (chevre). Results showed that the two conditions behaved similarly. In Experiments 2-4, we compared the intact diacritical words with a condition containing a mismatching diacritic, either existing in French (e.g., chévre, chevre) or not (the macron sign, as in chevre). We only found a reading cost when replacing the diacritic with an existing one. In Experiments 5-6, we compared the semantic categorization times to intact nondiacritical words (e.g., cheval, horse in English) versus a condition with an added diacritic, either existing (chèval) or not (cheval). We found a reading cost for the words with the added diacritical mark in both cases. We discuss how models of visual-word recognition can be modified to represent diacritical vowels.