Auditory brainstem responses (ABR) are used for both clinical and research purposes to objectively assess human hearing. A prominent feature of the transient evoked ABR is the level-dependent latency of the distinct peaks in its waveform. The latency of the most prominent peak, wave-V, is about 8 ms at a peak equivalent sound pressure level of 55 dB, and reduces for increasing level by approximately 1 ms / 20 dB. A classical explanation for this finding asserts that an increasing stimulus levels lead to a broadened excitation pattern on the basilar membrane. This results in further activation of the basal regions of the cochlea. Given the physical properties of the basilar membrane, increased basal activation is believed to cause a decreasing ABR latency. An Auditory Nerve (AN) model and the Dual Resonance Non-Linearity (DRNL) filter model are considered as separate front-end cochlear models to simulate ABRs. Even though both models incorporate level-dependent tuning and synapse adaptation, and thus theoretically should be capable of simulating level-dependent latencies, both models under-predict the latencies. The failure to produce accurate simulations suggests, that the level-depending tuning in the models is not accurately modelled. The level dependency of the basilar membrane filter tuning in humans is not well described in the literature and could therefore cause the modelling difficulties.