The purpose of this study was to define an optimal strategy for the tomographic reconstruction procedure in routine brain single-photon emission tomography (SPET) studies, including the number of projections, filter function and matrix size. A set of projection data with different count densities was obtained from a technetium-99m hexamethylpropylene amine oxime (99mTc-HMPAO) brain SPET acquisition from one volunteer. The projections were reconstructed with different filters and the quality of the reconstructed images was determined using both a subjective observer rating score and the Gilbert index. For each count density, the observers' choice corresponded to images with the lowest Gilbert index. The noise level in brain SPET sections was estimated and correlated with the fractal dimension. The results of this study indicate that although noise represents a fundamental component of brain SPET imaging, image quality also depends on the reconstructed spatial resolution. Image quality is satisfactorily described by fractal dimension. In addition the optimal filter function depends on the available count density. For high count levels, optimal reconstruction may be obtained by using a high-resolution matrix and a slightly smoother reconstruction filter. When count densities are low, best results are obtained by using a low-resolution matrix and a sharper filter. Finally, this study suggests that image quality is not influenced by the number of projections for equivalent count densities. These results were confirmed by 30 HMPAO brain SPET studies acquired in a routine clinical setting.