Learners detect probabilistic regularities in linguistic materials (Saffran & Kirham, 2018). In two experiments, we tested if this ability is active in reading and may underlie word learning. Italian speakers were shown letter strings and instructed to distinguish between novel words and foil distractors. Participants successfully detected that novel words complied with the statistics of Italian lexicon, and used this information to perform the task. Interestingly, among the several possible variables, they relied particularly on single bigram frequency. These data contrast with studies suggesting that readers average the frequencies across all bigrams in a string (Grainger et al., 2012), but accord with overall evidence for the role of discovering statistical structure in reading and language.