Word segmentation and word recognition were investigated by an examination of the word superiority effect in the context of a Chinese sentence. The same basic task was employed in four experiments, in which, on each trial, subjects searched for a predesignated character in a sentence. The target character was embedded in a two-character word, a two-character nonword, a string of asterisks, or a string of scrambled characters. Where there was a word or a nonword, it was manipulated to fit or not fit the meaning of the sentence. The results show that detection of a character was fastest when it was embedded in a string of asterisks, was faster when it was part of a word, whether or not the word fit the meaning of the sentence, and was slowest when it was part of a nonword or was among a string of scrambled characters. The word superiority effect was evident in the comparison of the word conditions with the scrambled-character condition. The presence of a word superiority effect, whether or not the word fit the meaning of the sentence, indicated that the sentence was not normally read; word segmentation and recognition nevertheless took place during character search. Further, such processes must have occurred fairly automatically, given that they were never the primary goal of the task.