Transient-evoked otoacoustic emissions (TEOAE) were used to characterize linear and nonlinear cochlear transduction using a new system-identification procedure. In this technique, a computational model of the system is first developed. From the measured stimulus and response records, spectral-density functions and multiple coherence functions are calculated. The coherence functions allow the characterization of linear/nonlinear processes as a function of frequency. Summations of linear and nonlinear coherences provide a goodness-of-fit of the chosen model. Finite impulse response pulses with a bandwidth of 1-8 kHz were used to evoke otoacoustic emissions. Eleven adults with normal hearing served as subjects. Third- and fifth-order polynomial models were used to model the data, and the results indicate that the fifth-order model is a better fit to the TEOAE data. The results of this study suggest that this system-identification procedure can be successfully applied to model cochlear transduction using a broadband stimulus. Most importantly, coherence functions provide useful insights into linear and nonlinear cochlear processes and have the potential to be developed as a clinical measure for monitoring changes in cochlear status.