Our brain is constantly performing predictions about upcoming events. By doing this, we can anticipate actions to take and rapidly respond to those probable future events. For example, during visual tasks, such as reading, predictions allow us to plan and perform ocular movements (that are responsible for the inspection and processing of the visual field) and to integrate words with their context more easily. In a series of studies, we analyzed the brain bases of predictions performed while reading natural stimuli, and modeled the Predictability (i.e. the variable that represents how probable is to guess a word before reading it) with computational models: (1) on an eye movement experiment, where participants read short stories, we analyzed how different computational models can mimic Predictability effect on gaze duration; (2) on an EEG experiment, where participants read Memory-Related and Common Sentences in a Serial Visual Presentation paradigm, we analyzed how different sources of predictions impacts on Evoked Potentials, like the N400; (3) finally, in an eye-movements and EEG co-registration study, where participants read Memory-Related and Common Sentences in a natural reading paradigm, we analyzed how Evoked Potentials are modified by natural reading. We conclude that predictions are performed using different sources (semantical, grammatical, syntactical, mnemonic, etc), and that there are different brain mechanisms underlying mnemonic predictions.