Ra-223, an α-particle emitting bone-seeking radionuclide, has recently been used in clinical trials for osseous metastases of prostate cancer. We investigated the relationship between absorbed fraction-based red marrow dosimetry and cell level-dosimetry using a model that accounts for the expected localization of this agent relative to marrow cavity architecture. We show that cell level-based dosimetry is essential to understanding potential marrow toxicity. The GEANT4 software package was used to create simple spheres representing marrow cavities. Ra-223 was positioned on the trabecular bone surface or in the endosteal layer and simulated for decay, along with the descendants. The interior of the sphere was divided into cell-size voxels and the energy was collected in each voxel and interpreted as dose cell histograms. The average absorbed dose values and absorbed fractions were also calculated in order to compare those results with previously published values. The absorbed dose was predominantly deposited near the trabecular surface. The dose cell histogram results were used to plot the percentage of cells that received a potentially toxic absorbed dose (2 or 4 Gy) as a function of the average absorbed dose over the marrow cavity. The results show (1) a heterogeneous distribution of cellular absorbed dose, strongly dependent on the position of the cell within the marrow cavity; and (2) that increasing the average marrow cavity absorbed dose, or equivalently, increasing the administered activity resulted in only a small increase in potential marrow toxicity (i.e. the number of cells receiving more than 4 or 2 Gy), for a range of average marrow cavity absorbed doses from 1 to 20 Gy. The results from the trabecular model differ markedly from a standard absorbed fraction method while presenting comparable average dose values. These suggest that increasing the amount of radioactivity may not substantially increase the risk of toxicity, a result unavailable to the absorbed fraction method of dose calculation.