We present three-dimensional simulations of the image formation process in near-field optical microscopy, Our calculations take into account the different components of a realistic experiment: an extended metal coated tip, a subwavelength sample and its substrate. Mie investigate all possible detection (transmitted, reflected and collected field) and scanning (constant height, constant gap) modes. Our results emphasize the strong influence of the tip motion on the experimental signal. They also show that it is possible, by controlling the polarization of both the illumination and the detected field, to strongly reduce these artefacts.