Our sensory systems take in vast amounts of information about our external surroundings. We would certainly be overwhelmed by this if not for the ability to form internal predictions about how external stimuli in our near future will look, sound, smell, etc. To form these predictions, neural systems are engaged at different stages. One theory is that this is a linear process, meaning that at each stage these systems process and evaluate the predictions drawn by prior stages. However, this theory does not explain what might occur if predictions made by previous stages contradict each other. In this issue, Tabas and Kriegstein investigated how expectations at each stage of the sensory processing pipeline are used to form internal predictions. They measured fMRI responses from the human auditory cortex, thalamus, and midbrain as participants listened to repetitive sequences of sounds. Participants made predictions on when “deviant” tones occurred in each sound sequence based on either a strategy supporting the linear processing theory, a strategy supporting multiple stages being engaged to process a prior stage, or both. The authors found that populations of neurons in all brain regions of interest required both types of predictions. This suggests that internal predictive coding engages with the auditory system in a nonlinear way. These data advance our understanding of how complex auditory signals, like speech, are sensed and understood and are informative for future studies on the auditory system and predictive coding.