Ninetieth percentile cutoffs for acoustic reflex thresholds (ARTs) were determined for a sample of 2,748 ears of 1,374 subjects with normal hearing and sensorineural loss of cochlear origin. All subjects had measurable hearing (less than or equal to 110 dB HL, ANSI-1969) at all three activator frequencies (500, 1000, and 2000 Hz). Cutoff values including "no responses" ("absent" reflexes at 125 dB HL) were higher than those excluding no responses when hearing losses were greater than about 55 dB. The 90th percentiles including the effects of no responses identified ears with retrocochlear involvement for hearing losses as great as about 756 dB. For greater hearing losses at the activator frequency, the no-response rate for both cochlear and retrocochlear cases is too high to enable them to be differentiated by acoustic reflex thresholds. The 90th percentiles are derived at each activator frequency collapsed across ears. It is therefore necessary to determine the probabilities that normal or cochlear-impaired ears will have one, two, or three frequencies at which the ARTs exceed their respective 90th percentiles. It was found that among normal and cochlear-impaired ears, 12.2% have one ART elevated above the 90th percentile, but only 5.6% have two or three elevated ARTs. Clinical implications are discussed.