The experiments reported here examine the psychoacoustical bases for poor speech perception in noise by persons with sensorineural impairment. Two major hypotheses are tested. First, because persons with cochlear impairment are less able than normal-hearing persons to separate out incoming signals on the basis of spectral differences (a deficit referred to as reduced frequency selectivity), they are less able to localize one sound in the presence of other sounds. Second, this reduced localization ability makes it difficult for the hearing-impaired person to take advantage of the spatial separation of a target speech source and other interfering sources. Such separation is common in real environments and facilitates speech perception by normal-hearing persons. Tests of these hypotheses are conducted by means of detailed psychoacoustical measures of frequency selectivity, of localization and speech perception under masking.