An image recording apparatus in which at least one light beam scans over an inner surface of a cylindrical member to record an image on a recording medium held on the inner surface. A light beam from light beam outputting means advances along a central axis of a cylindrical member and enters an optical head which includes deflecting means and an imaging optical system. The deflecting means and the imaging optical system rotate as one unit. The nodal point of the imaging optical system is located at or close to a point on the central axis which is nearest a beam irradiation position on the recording medium. Hence, displacement of the imaging position is prevented which makes it possible to record a high quality image, even if the optical head wobbles.