Word repetition has been a staple paradigm for both psycholinguistic and memory research; several possible loci for changes in behavioral performance have been proposed. These proposals are discussed in light of the event-related brain potential (ERP) data reported here. ERPs were recorded as subjects read nonfiction articles drawn from a popular magazine. The effects of word repetition were examined in this relatively natural context wherein words were repeated as a consequence of normal discourse structure. Three distinct components of the ERP were found to be sensitive to repetition: a positive component peaking at 200 msec poststimulus, a negative one at 400 msec (N400), and a later positivity. The components were differentially sensitive to the temporal lag between repetitions, the number of repetitions, and the normative frequency of the eliciting word. The N400 responded similarly to repetition in text as it has in experimental lists of words, but the late positivity showed a different pattern of results than in list studies.