The radionuclides used in nuclear medicine imaging emit numerous mono-energetic electrons responsible for dose heterogeneity at the cellular level. S(self) the self-dose per unit cumulated activity (which results from the radionuclide located in the target cell), and S(cross) the cross-dose per unit cumulated activity (which comes from the surrounding cells) delivered to a target cell nucleus by electron emissions of technetium-99m, iodine-123, indium-111, gallium-67 and thallium-201 were computed at the cellular level. An unbounded close-packed hexagonal cell arrangement was assumed, with the same amount of radioactivity per cell. Various cell sizes and subcellular distributions of radioactivity (nucleus, cytoplasm and cell membrane) were simulated. The results were compared with those obtained using conventional dosimetry. S(self) and S(cross) values depended closely on cell dimensions. While the self-dose depended on the tracer distribution, the latter affected the cross dose by less than 5%. When the tracer was on the cell membrane, the self-dose was particularly low compared to the cross-dose, as the self-dose to cross-dose ratio was always less than 11%. In the case of cytoplasmic or cell membrane distribution of radioactivity, conventional electron dosimetry slightly overestimated the dose absorbed by the target cell nucleus (by 1.08-to 1.7-fold). In contrast, conventional dosimetry strongly underestimated the absorbed dose (1.1- to 75-fold) when the radioactivity was located in the nucleus. The discrepancies between conventional and cellular dosimetry call for calculations at the cellular level for a better understanding of the biological effects of radionuclides used in diagnostic imaging.