Most classical models of visual word recognition are based on sequentially organized levels of representation and involve feedback mechanisms to various extents. In this study, we aim at clarifying which of the early processing stages of visual word recognition are modulated by top-down lexical effects. We studied the identification of letters embedded in briefly presented words (e.g., TABLE) and illegal nonwords (e.g., GTFRS) using event-related potentials (ERPs). Participants were involved in the Reicher-Wheeler paradigm: they were asked to indicate which of two letters displayed above and below a string of hashes was flashed immediately before at fixation within a letter string, which was either a word or a nonword. Event-related potentials were significantly modulated by the lexical status of stimuli around 200 ms after stimulus onset, i.e., in the peaking window of the N1 component. In light of our results, we propose that visual word form representations can constrain letter identification at a prelexical stage i.e., during the extraction of letter-shape information. In addition, we show that this facilitatory top-down effect is sensitive to stimulus exposure duration.