The study aimed at the development of a clinically applicable methodology that could: (1) discriminate transient evoked otoacoustic emission (TEOAE) recordings from normal hearing or hearing impaired individuals; (2) classify the nature of the hearing loss as conductive or as cochlear, and (3) define clear-cut TEOAE clinical criteria. A classification algorithm based on a multivariate discriminant analysis of fast Fourier transform data from recordings evoked by click stimuli of 50 +/- 2, 62 +/- 2, 68 +/- 2 and 80 +/- 2 dB SPL was used to discriminate 302 normal subjects from 383 subjects suffering from mild to moderate hearing losses. The best discriminant model (QDF80) produced a sensitivity of 93.8% and a specificity of 79.4%. When extra correlation criteria were serially applied to the classification outcome, the specificity was increased to 85.3%, but the sensitivity was marginally decreased to 91.7%. The classification of the correctly identified hearing-impaired cases yielded 93.8% identification of conductive and 75.1% identification of cochlear cases. A sensitivity analysis of the misclassified hearing-impaired cases suggested that the TEOAE spectra are well correlated with the 2-kHz but poorly correlated with the 4-kHz octave frequency.