The goal of this manuscript was to give a short overview of the current knowledge of the neural correlates of the late auditory evoked potentials and to interpret these observations in the concept of the Bayesian brain, which we further applied to the currently limited results in tinnitus patients. Based on these interpretations, we propose some further studies and caveats for AEP studies in tinnitus. We mainly focused on the sequentially activated brain areas in auditory processing by making use of EEG or MEG measurement. As these techniques have a rather low spatial resolution and the identification of subcortical regions is still a matter of debate, results should be interpreted carefully. Although notion should be made that many of these regions are identified by intracerebral recordings or fMRI, as well. Based on the current results, it should be said that there is no clear cut-off between bottom-up and top-down processing; it should rather be seen as a transitional, or interactive, process. This corresponds to the Bayesian predictive brain model. If the brain is truly predictive, there should be brainstem evoked potentials that are already influenced by top-down processes even before perception is possible, i.e. filtering auditory input at an early stage, compatible with a Bayesian model of information processing, as has been proposed for the visual system (Lee and Mumford, 2003). The FFR can indeed be attentionally modulated, even at the level of the brainstem (Du et al., 2012). In conclusion, future research making use of AEPs could be of great interest to further unravel the fundamental neurobiological mechanisms of tinnitus and auditory perception. But, future studies should take into account that the presented tones should be adjusted to the individual tinnitus pitch and frequency region of hearing loss, one of the most prominent shortcomings of the currently published studies. In addition, it would be of great interest to not only look at the AEP characteristics, but also at different composing frequencies.