For over a decade, it has been known that intended actions and their sensory consequences are judged as occurring closer together in time than otherwise equivalent non-intended actions and their effects. This phenomenon, known as temporal binding, has been argued to represent a sense of agency and to be the result of motor-predictive mechanisms underlying voluntary action. It has further been demonstrated that such binding can be affected by learning. Based on these findings, I here build on the motor-predictive framework and complementary theories of learning and argue for a central role for prediction error in temporal binding. A computational framework for estimating on-line task-irrelevant predictions in a binding task is proposed and tested on 17 participants while they are simultaneously undergoing functional magnetic resonance imaging (fMRI). It was hypothesised that predictions would be positively associated with degree of binding while the actual outcome would interact with the prediction and form a negative association in terms of increased prediction error. I further speculated that the degree of prediction error would be represented by neural activity in areas more commonly associated with reinforcement learning. The behavioural analysis showed that prediction error was significantly predictive of decreased binding, as hypothesised, but no significant independent contribution of the prediction regardless of outcome was found. For the fMRI analysis, no significant activation was found to be related to prediction error or the prediction itself. However, a second analysis looking for correlations with binding measures revealed negatively correlated activation in the precuneus and left upper brain-stem. The results are discussed in light of the binding measure currently used and a potential role for the default-mode network in agency. It is concluded that task-irrelevant learning of stimulus-identity seems to occur in temporal binding, and seems driven by prediction error. It is, however, still unknown how this is implemented in neural terms. Because the current study fails to replicate an earlier finding regarding neural correlates of temporal binding, a continued investigation of this is encouraged, as well as further studies on how different binding measures relate to agency and voluntary action.