Subjects of the present experiment were asked to read different sentences and to respond to occurrence of a given character for each sentence. In each trial, a sentence was displayed either as a whole on a computer screen or as successive presentations of single characters toa common location. Results showed a reliable word superiority effect: Latencies were significantly longer and detection errors greater when a given character was embedded in a nonword rather than in a word context. This word superiority effect for errors diminished when the sentences were presented in the one-character condition. In addition, fewer errors were made when the sentences were displayed in the one-character rather than in the whole sentence condition. These findings are discussed in terms of how people process characters and words in reading situations.