In German, diacritical marks distinguish between vowel sounds in print (e.g., “o” pronounced /o/ vs. “ö” pronounced /ø/). Unlike in Spanish, where diacritics primarily indicate lexical stress, omitting diacritics in German (e.g., Kröte [toad] → Krote) leads to longer word identification times compared with intact words. This suggests separate letter representations for diacritical and nondiacritical vowels in German. Current models of visual word recognition assume distinct letter representations for diacritical and nondiacritical vowels in German (Ziegler et al., 2000), but it remains unclear whether the reading cost differs when a diacritic is added versus omitted. We conducted three semantic categorization experiments to examine whether the presence of an added diacritic in a nondiacritical word (e.g., Schwan [swan] → Schwän) incurs a greater lexical-semantic cost than its omission (e.g., Kröte → Krote) in German and Finnish, another language where diacritical vowels signal distinct pronunciations. In noisy-channel models, adding a diacritic makes the percept less similar to the base word than omitting one, thus predicting a larger cost. In contrast, abstractionist models assume rapid activation of abstract letter representations, predicting a negligible asymmetry. Results were similar in German and Finnish. First, both types of misspellings showed a reading cost relative to the intact words. Second, the reading cost was larger for the addition than for the omission of diacritics, placing new constraints on the orthographic front-end of models of visual word recognition.