Onomatopoeia is widespread across the world’s languages. They represent a relatively simple iconic mapping: the phonological/phonetic properties of the word evokes acoustic related features of referents. Here, we explore the EEG correlates of processing onomatopoeia in English. Participants were presented with a written cue-word (e.g., leash) and then with a spoken target-word. The target-word was either an onomatopoeia (e.g., bark), a sound-related but arbitrary word (e.g., melody), or another arbitrary word (e.g., bike). Participants judged whether the cue- and the target-word were similar in meaning. We analysed Event-Related Potentials (ERPs) in different time-windows: (i) early (100–200 and 200–250 ms) to assess differences in processing at the form-level; (ii) the N400 time-window (300–500 ms) in order to establish if there are differences in semantic processing across our word-types; and (iii) late (600–900 ms) to assess post-lexical effects. We found that onomatopoeia differed from the other words in the N400 time-window: when cue and target were unrelated, onomatopoeic words led to greater negativity which can be accounted for in terms of enhanced semantic activation of onomatopoeia which leads to greater salience of the mismatch. We discuss results in the context of a growing body of literature investigating iconicity in language processing and development.