Two experiments tested memory for a once-heard voice. Experiment 1 showed that hearing an unknown voice utter the same sentence three times improved recognition of the voice a week later compared with hearing only a single utterance of the sentence, but instructions to attend to the voice did not. The impairment caused by seeing the speaker's face during the original utterance (the Face Overshadowing Effect--FOE) occurred only with the single utterance and was unaffected by instructions. It was concluded that length of utterance without increase in speech variety aids voice recognition, and that the FOE is not due to voluntary attention to face rather than voice, but may be due to involuntary attention to the initial exposure of the face. Experiment 2 investigated whether exposure to the face before the onset of speech reduced the FOE. Such an effect was observed and it was concluded that the FOE is due to involuntary preference for processing face information in order to aid later recognition of the individual, while speech is processed primarily for meaning. Absence of an FOE on memory for the words uttered was consistent with this interpretation.