Numerous studies have provided evidence that readers generate phonological codes while reading. However, a central question in much of this research has been how early these codes are generated. Answering this question has implications for the roles that phonological coding might play for skilled readers, especially whether phonological codes affect the identification of most words, which can only be the case if these codes are generated rapidly. To investigate the time course of phonological coding during silent reading, the present series of experiments examined survival analyses of first-fixation durations on phonologically related (homophones, pseudohomophones) and orthographic control (orthographically matched words and nonwords) stimuli that were either embedded in sentences in place of correct targets (Experiments 1 and 2) or presented as parafoveal previews for correct targets using the boundary paradigm (Experiments 3 and 4). Survival analyses revealed a discernible difference between processing the phonologically related versus the orthographic control items by as early as 160 ms from the start of fixation on average (160-173 ms across experiments). Because only approximately 18% of first fixation durations were shorter than these mean estimates and follow-up tests revealed that earlier divergence point estimates were associated with shorter gaze durations (e.g., more rapid word identification), results suggest that skilled readers rapidly generate phonological codes during normal, silent reading and that these codes may affect the identification of most words.