Internal dosimetry deals with the determination of the amount and the spatial and temporal distribution of radiation energy deposited in tissue by radionuclides within the body. Nuclear medicine has been largely a diagnostic specialty, and model-derived average organ dose estimates for risk assessment, the traditional application of the MIRD schema, have proven entirely adequate. However, to the extent that specific patients deviate kinetically and anatomically from the model used, such dose estimates will be inaccurate. With the increasing therapeutic application of internal radionuclides and the need for greater accuracy, radiation dosimetry in nuclear medicine is evolving from population- and organ-average to patient- and position-specific dose estimation. Beginning with the relevant quantities and units, this article reviews the historical methods and newly developed concepts and techniques to characterize radionuclide radiation doses. The latter include the 3 principal approaches to the calculation of macroscopic nonuniform dose distributions: dose point-kernel convolution, Monte Carlo simulation, and voxel S factors. Radiation dosimetry in "sensitive" populations, including pregnant women, nursing mothers, and children, also will be reviewed.