Most speech research with infants occurs in quiet laboratory rooms with no outside distractions. However, in the real world, speech directed to infants often occurs in the presence of other competing acoustic signals. To learn language, infants need to attend to their caregiver's speech even under less than ideal listening conditions. We examined 7.5-month-old infants' abilities to selectively attend to a female talker's voice when a male voice was talking simultaneously. In three experiments, infants heard a target voice repeating isolated words while a distractor voice spoke fluently at one of three different intensities. Subsequently, infants heard passages produced by the target voice containing either the familiar words or novel words. Infants listened longer to the familiar words when the target voice was 10dB or 5 dB more intense than the distractor, but not when the two voices were equally intense. In a fourth experiment, the assignment of words and passages to the familiarizationand testing phases was reversed so that the passages and distractors were presented simultaneously during familiarization, and the infants were tested on the familiar and unfamiliar isolated words. Duringfamiliarization, the passages were 10 dB more intense than the distractors. The results suggest that this may be at the limits of what infants at this age can do in separating two different streams of speech. In conclusion, infants have some capacity to extract information from speech even in the face of a competing acoustic voice.As adults, we are constantly faced with the problem of listening to one speaker amidst various other noises. Often, we have to deal with competition from other speakers as well. This is classically known as the cocktail party effect-an allusion to our ability to follow a conversation even in the midst ofa party environment, when many people are speaking simultaneously in a relatively small area ofspace. This type ofselective attention is related to a more general phenomenon known as streaming, which refers to our ability to group together sounds originating from one source and separate them from sounds originating from other sources.The cocktail party effect and streaming have been the subject ofextensive research since the 1950s. Listeners are known to separate different sound streams on the basis of a variety ofacoustic cues, such as location in space (Broadbent,