Various methods, such as those developed by the Medical Internal Radiation Dosimetry (MIRD) Committee of the Society of Nuclear Medicine or employing dose point kernels, have been applied to the radiation dosimetry of (131)I radionuclide therapy. However, studies have not shown a strong relationship between tumour absorbed dose and its overall therapeutic response, probably due in part to inaccuracies in activity and dose estimation. In the current study, the GATE Monte Carlo computer code was used to facilitate voxel-level radiation dosimetry for organ activities measured in an (131)I-treated thyroid cancer patient. This approach allows incorporation of the size, shape and composition of organs (in the current study, in the Zubal anthropomorphic phantom) and intra-organ and intra-tumour inhomogeneities in the activity distributions. The total activities of the tumours and their heterogeneous distributions were measured from the SPECT images to calculate the dose maps. For investigating the effect of activity distribution on dose distribution, a hypothetical homogeneous distribution of the same total activity was considered in the tumours. It was observed that the tumour mean absorbed dose rates per unit cumulated activity were 0.65E-5 and 0.61E-5 mGY MBq(-1) s(-1) for the uniform and non-uniform distributions in the tumour, respectively, which do not differ considerably. However, the dose-volume histograms (DVH) show that the tumour non-uniform activity distribution decreases the absorbed dose to portions of the tumour volume. In such a case, it can be misleading to quote the mean or maximum absorbed dose, because overall response is likely limited by the tumour volume that receives low (i.e. non-cytocidal) doses. Three-dimensional radiation dosimetry, and calculation of tumour DVHs, may lead to the derivation of clinically reliable dose-response relationships and therefore may ultimately improve treatment planning as well as response assessment for radionuclide therapy.