This paper addresses the issue of numerical refocusing and its limits using the linear systems theory.Numerical refocusing is a well-known technique allowing looking above and below the focal plane of a sample in a manner similar to mechanically refocusing, but it can be used off-line. The most common implementation of this method is multiplication of angular spectrum of the recorded complex amplitude by the Fresnel propagating function. However, this approach is correct only in the case when monochromatic and spatially coherent illumination is used. Therefore it is not generally usable. If we concern ourselves with planar samples, we can afford to go into more details and we can perform the refocusing correctly, providing that we know imaging properties of the system in form of its transfer function. For instance, if we record the wave at a certain position, we can refocus as far as we want without introducing any error of this kind. Moreover, if we work with a binary approximation of the transfer function, we can derive a more general propagator that can be used to refocus under incoherent illumination with much more accurate results.© (2010) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.