Our own actions always precede subsequent sensory feedback through which we can regulate the action. This temporal relationship between our own actions and the associated sensory feedback provides a clue for perceiving causality and discriminating the sensory feedback from other irrelevant sensory events. However, the judgment of the temporal order of an own action and a sensory event might be confounded, as the physical and neural delays of the sensory information are not accessible. This issue can be solved by statistical learning; that is, if a particular temporal difference between an action and a sensory event is very frequent, that event is likely to be the feedback. To prove the statistical learning hypothesis, we manipulated the temporal difference between an action and the sensory feedback. In this experiment, each session began with an adaptation phase during which the participants’ voluntary button presses were coupled with noise bursts, and the actions and noise bursts were separated by a fixed time lag of -15, 0, 100, or 200 ms (the negative value indicates that the noise burst occurred first). Following the initial adaptation, the participants were asked to press the button again, and a noise burst was presented at one of several possible onset asynchronies relative to the button press. The participants made either ‘sound-first’ or ‘tap-first’ responses. The results showed that the point of subjective simultaneity shifted as a function of the adapted asynchronies, including the situation in which the feedback preceded the action, suggesting that the recalibration process is flexible. Recalibration indeed occurs, supporting the statistical learning hypothesis.