A current value of an intake valve phase calculated based on a sensor output is smoothed along the direction of time axis in accordance with a smoothing factor, whereby a phase detection value is detected. At the time of reference position learning, the smoothing factor is set to a value larger than at the time of normal control, so that the degree of smoothing in the smoothing process of phase detection value becomes smaller. Therefore, it is possible to detect more quickly that the intake valve timing has reached the reference timing at the time of reference position learning based on the phase detection value, than when a common smoothing factor is set both for the normal control and for the reference position learning.