Event-related potentials (ERPs) were recorded during recognition tasks for spoken words alone (items) or for both words and the voice of the speaker (sources). Neither performance nor ERP measures suggested that voice information was retrieved automatically during the item-recognition task. In both tasks, correctly recognized old words elicited more positive ERPs than new words, beginning around 400 ms poststimulus onset. In the source task only, old words also elicited a focal prefrontal positivity beginning about 700 ms. The prefrontal task effect did not distinguish trials with accurate and inaccurate voice judgments and is interpreted as reflecting the search for voice information in memory. More posterior recording sites were sensitive to the successful recovery of voice or source information. The results indicate that word and voice information were retrieved hierarchically and distinguish retrieval attempt from retrieval success.