The question as to how accurately small object features can be reproduced in optical microlithography does not have a simple answer. It depends not only on the dimensions of the feature, but also on whether it is a line or a space, whether there are other features nearby (the proximity effect), on the resist thickness, and on whether the features are to have the dimensions of the ideal optical image or are "biased". This paper explores these topics by modeling the imaging, exposure, and development steps. In order to discover the significant dependencies we first investigate the simple model of a resist layer deposited onto a non-reflecting substrate. It shows that an isolated resist line has approximately the correct dimensions in all sizes when the imaging is done with partially coherent light. The isolated spaces, however, show deviations from the design size, which are caused by diffraction effects. For spaces (0.6 - 1.0) λ/NA wide the light intensity in the center of the space is larger than the intensity of the incident light causing the resist to develop through faster than in a very large area For spaces smaller than 0.6 λ/NA the light intensity in the image drops rapidly and the spaces can no longer be reproduced. The dependence on coherence of the light, on resist thickness, and on the degree of focus are also investigated. On real sur-faces of silicon, Si02, and aluminum reflections cause interference effects and dramatic variations in exposure with resist thickness. After these effects are taken into account, the resolution is much poorer (0.9 λ/NA in the case of silicon, worse for aluminum). Biasing of the masks does not improve the resolution capability. On the other hand, a post-bake of the re-sist pattern can cause a dramatic improvement of the imaging quality and can increase the resolution to where it is comparable to that obtained on the non-reflective substrate.