It is clear that the representation of complex stimuli, such as speech, is tonotopic at peripheral levels of the auditory system. That is, the array of auditory nerve fibers are arranged according to their frequency sensitivity, and a particular fiber informs the CNS about the acoustic stimulus at frequencies near the fiber's best frequency (BF). The nature of the representation at a particular frequency is still unclear, however. According to the rate-place model a fiber signals the energy at its BF by changing its discharge rate; the alternative is that the representation depends on the detailed temporal patterns of response of the fibers. It has long been known that the rate-place code is significantly less robust in the face of high sound levels and background noise than is the temporal representation. Although this result suggests that the brain might use information encoded in the temporal aspects of response, there are difficulties in understanding how the CNS would extract temporal information; moreover, temporal information only exists up to 4-6 kHz, the upper limit of the phase-locking on which the temporal code depends. Yet animals, including humans, clearly use information encoded in the stimulus spectrum at higher frequencies, in localizing sound for example. In this talk, the results of a more detailed examination of the rate code in the auditory nerve and cochlear nucleus will be described. We show that the rate code is more robust than previously thought, that cats' ability to discriminate vowel second formant frequency corresponds closely to predictions derived from the rate code, and that cochlear nonlinearities maintain rate sensitivity in complex stimuli by maintaining sharp tuning over a wide range of sound levels. Despite all this, there are some situations, such as separation of multiple speech streams, where temporal information is probably used.