With the aim of identifying event-related potential (ERP) correlates of the tip-of-the-tongue (TOT) state, we employed a task involving identification and naming of the faces of famous people with three response categories: KNOW (correct naming), DON'T KNOW (name not known), and TOT. Before 450 ms, no differences are observed between TOT and KNOW conditions in ERPs related to face processing and recognition. The smaller mean amplitude in TOT than in KNOW responses over the 550-750-ms interval and the N-TOT difference wave are correlates of the onset of the TOT state, probably reflecting a division of processing resources and low activation of the lexical-phonological route. The smaller amplitude of a late negative wave (between 1350 and 1550 ms) in TOT than in DON'T KNOW responses, and the P-TOT difference wave, seem to indicate that an important amount of processing resources is engaged in resolving the TOT.