A comprehension-sensitive proofreading task was used to examine phonological recoding during the silent reading of natural text. In Experiment 1, adult readers identified inconsistent words that were or were not homophones of consistent ones; phonology was implicated by the finding that readers were less likely to detect the inconsistent word in You vile villain to steel my gold! than in You vile villain to stale my gold!, but only if exposed to an error-free version first. Experiments 2 and 3 ruled out working memory and word-specific priming explanations, which suggests that the homophone effect reflected phonological codes activated during lexical access. Although homophony interfered with the initial detection of an inconsistent word, Experiment 4 showed that it facilitated the recovery of the correct solution (e.g., steal). We interpret the homophone effects within the framework of a context-sensitive version of Van Orden's (1987) lexical access model.Consider the task of having to read the following prose passage while proofreading it for errors:Russell Wood, a nearly bald man of modest means, wore clothes of worn denim and a tie with flecks of color. Unless you focused on his tie, he almost disappeared. His hair matched his face, which matched his suit-all beige.Alone at his teller's cage, idle and board, Russell gazed at the frieze along the bank's ceiling. He swayed back and forth and the room began to spin as he daydreamed Now consider reading and proofreading the following version: Russell Wood, a nearly bald man of modest means, wore clothes of worn denim and a tie with flecks of color. Unless you focused on his tie, he almost disappeared. His hair matched his face, which matched his suit-all beige.Alone at his teller's cage, idle and beard, Russell gazed at the frieze along the bank's ceiling. He swayed back and forth and the room began to spin as he daydreamed Would you predict a difference in the detectability of the error board in the first version versus the error beard in the second? Both of these words are semantically inconsistent in the context of the passage. However, board happens to sound identical to the word bored, which is perfectly consistent in that context, whereas beard does not. If proofreaders were more likely to miss an error that was a homophone of the correct word, this might suggest that readers translate orthographic representations to phonological representations when comprehending printed text. By translating the orthographic representation board Xo its phonological representation /bord/ This research was supported by a grant from the Natural Sciences and Engineering Research Council of Canada to Meredyth Daneman.We thank Joan Collins, Alda Fernandes, and Toni Marlow for their help in data collection. We also thank Morris Moscovitch for his comments and suggestions, and Derek Besner, Betty Ann Levy, and two anonymous reviewers for their comments on an earlier version.