The invention relates to a method of reconstructing a high-resolution 3D image (G) of an examination zone of a patient from a 3D image data set (D) of the examination zone, the examination zone being subject to a periodic motion which is measured, in parallel with the acquisition of the 3D image data set (D), as a motion signal (E) which represents the periodic motion. In order to enable the formation of high-resolution 3D images having an enhanced image quality, the invention proposes a method of this kind which includes the following steps: a) reconstructing a number of low-resolution 3D images (I) from the 3D image data set (D), the low-resolution 3D images (I) being reconstructed from 3D image data of the 3D image data set (D) which has been acquired during different phases of motion of the periodic motion, b) determining motion information (B) of at least one sub-zone (A) of the examination zone during the different phases of motion by means of the low-resolution 3D images (I), c) selecting a temporal reconstruction window (T) in which the motion of the at least one sub-zone (A) is below a predetermined level, d) reconstructing a high-resolution sub-image (K) for the at least one sub-zone (A) from 3D image data (D) lying in the temporal reconstruction window (T) selected for the sub-zone (A), and e) forming the desired 3D image (G) from the at least one sub-image (K), the zones of the 3D image (G) which are not reconstructed as sub-images then being reconstructed from the 3D image data so as to be combined with the at least one sub-image (K).