High resolution electron microscope images showing the detailed distribution of metal atoms within the unit cells of complex oxide structures have been recorded recently and as a first approximation may be interpreted as amplitude-object images if obtained with the degree of defocus corresponding to the "optimum-defocus condition" for the phase-contrast imaging of thin phase objects. Detailed observations of images of Ti2Nb10O29 crystals having thicknesses of the order of 100 Å reveal that the thin phase-object approximation, which assumes that only small phase-shifts are involved, is inadequate to explain some features of the image intensities including the variation of contrast with crystal thickness. A very aproximate treatment of the phase contrast due to defocussing of phase objects having large phase shifts is evolved and shown to give a qualitativity correct account of the observations. The variation of image contrast with tilt away from a principle orientation is discussed. From the symmetry of the image contrast it is deduced that the symmetry of the crystal structure as derived from X-ray diffraction studies can not be correct.