The resolution properties of images created via maximum-likelihood reconstruction are affected by the recon struction parameters, and data and object characteristics as well, due to the non-linear nature of the maximum-likelihood (ML) algorithm. In most cases, the influence of these factors cannot be estimated analytically. We have studied these effects in the 3D TOF PET (three-dimensional time-of-flight positron emission tomography) statistical reconstructions by evaluating point spread functions (PSF) in the reconstructed images for simulated point sources positioned within uniform backgrounds of various properties (size, activity, attenuation) and employing various parameters and detector's line-of-response (LOR) res olution models within maximum-likelihood reconstructions. We have evaluated the shape and the full width at half maximum (FWHM) of the reconstructed point source images, convergence speeds, and stability of the reconstructions. Among the most important influences affecting the resolution and convergence speed of the reconstruction have been found to be: the voxel size, the local contrast level near the point source and the width of the LOR resolution model used in the reconstruction. Furthermore, reconstructions of point sources in a non-zero background (that is, not in air) take a very long time to converge to their ultimate shape, and especially so if utilizing resolution modeling within the reconstruction. The main contribution of this work is a systematic study of various effects and influences on the properties (mainly convergence speed and FWHM) of the resolution functions obtained via ML reconstruction. Results of this study provide a better understanding of individual influences on the ML reconstruction resolution properties and provide guidance for development of image-based resolution modeling.