Distortion product otoacoustic emissions (DPOAE) were measured in normal-hearing and hearing-impaired human subjects. Analyses based on decision theory were used to evaluate DPOAE test performance. Specifically, relative operating characteristic (ROC) curves were constructed and the areas under these curves were used to estimate the extent to which normal and impaired ears could be correctly identified by these measures. DPOAE amplitude and DPOAE/noise measurements were able to distinguish between normal and impaired subjects at 4000, 8000, and, to a lesser extent, at 2000 Hz. The ability of these measures to distinguish between groups decreased, however, as frequency and audiometric criterion used to separate normal and hearing-impaired ears decreased. At 500 Hz, performance was no better than chance, regardless of the audiometric criterion for normal hearing. Cumulative distributions of misses (hearing-impaired ears incorrectly identified as normal hearing) and false alarms (normal-hearing ears identified as hearing impaired) were constructed and used to evaluate test performance for a range of hit rates (i.e., the percentage of correctly identified hearing-impaired ears). Depending on the desired hit rate, criterion values of -5 to -12 dB SPL for DPOAE amplitudes and 8 to 15 dB for DPOAE/noise accurately distinguished normal-hearing ears from those with thresholds greater than 20 dB HL for the two frequencies at which performance was best (4000 and 8000 Hz). It would appear that DPOAE measurements can be used to accurately identify the presence of high-frequency hearing loss, but are not accurate predictors of hearing status at lower frequencies, at least for the conditions of the present measurements.