Action perception and execution are linked in the human motor system, and researchers have proposed that this action-observation matching system underlies our ability to predict observed behavior. If the motor system is indeed involved in the generation of action predictions, activation should be modulated by the degree of predictability of an observed action. This study used EEG and eye-tracking to investigate whether and how predictability of an observed action modulates motor system activation as well as behavioral predictions in the form of anticipatory eye-movements. Participants were presented with object-directed actions (e.g., making a cup of tea) consisting of three action steps which increased in their predictability. While the goal of the first step was ambiguous (e.g., when making tea, one can first grab the teabag or the cup), the goals of the following steps became predictable over the course of the action. Motor system activation was assessed by measuring attenuation of sensorimotor mu- and beta-oscillations. We found that mu- and beta-power were attenuated during observation, indicating general activation of the motor system. Importantly, predictive motor system activation, indexed by beta-band attenuation, increased for each action step, showing strongest activation prior to the final (i.e. most predictable) step. Sensorimotor activity was related to participants' predictive eye-movements which also showed a modulation by action step. Our results demonstrate that motor system activity and behavioral predictions become stronger for more predictable action steps. The functional roles of sensorimotor oscillations in predicting other's actions are discussed.