For calculation of radiation dose to the marrow, standard dosimetry for radiopharmaceuticals that do not bind to the marrow includes dose contributions from radioactivity in blood and the remainder of the body. For a pure β-emitter such as 90Y, marrow dose is usually determined by the blood contribution. However, myelotoxicity from 90Y-antibody therapy often correlates poorly with marrow dose estimated using the blood method. This study proposes a method to address 2 possible factors affecting marrow dose estimates. These include (a) recycled 90Y in bone/ marrow space after 90Y-antibody has been processed in the liver and (b) use of the marrow mass of Reference Man for individual patients. Methods: Thirty-three patients with advanced non-small cell lung cancer were treated with 90Y-anti-TAG-72 murine antibody (CC49). TAG-72 is often expressed in epithelial-derived tumors but not in normal marrow. 111In-CC49 was used as a tracer. The marrow doses from blood were calculated on the basis of radioactivity concentrations in blood. Marrow dose in the lumbar vertebrae was estimated from images for 111In-CC49 uptake in L2-L4. In 20 patients who had CT images, trabecular bone volumes of L2-L4 were measured from CT images to estimate patient-specific marrow mass in L2-L4. The fraction of baseline platelet counts at nadir was used as an indicator of myelotoxicity. Results: Marrow dose per unit injected radioactivity estimated from blood was lower than that from L2-L4 uptake values. Prediction of myelotoxicity using marrow dose estimated from blood was poorer than that using injected dose per body surface area (GBq/m2) (r = 0.31 vs. 0.51). Prediction was improved using marrow dose estimated from L2-L4 uptake, assuming the marrow mass of Reference Man (r = 0.67 for n = 33; r = 0.70 for n = 20). Prediction was worse if reference marrow mass was adjusted by body weight (r = 0.56 for n = 33; r = 0.63 for n = 20). Prediction was not improved if adjusted by body surface area or lean body mass but was improved if adjusted by height (r = 0.72 for n = 33; r = 0.78 for n = 20). The best prediction was obtained (r = 0.85 for n = 20) using patient-specific L2-L4 marrow mass estimated from CT. Conclusion: Marrow dose estimated from the blood radioactivity method was not a good predictor of myelotoxicity for non-marrow-targeting 90Y-antibody therapy. Thrombocytopenia in this group of patients correlated much better with dose estimated from lumbar vertebrae imaging and patient-specific marrow mass than with that estimated from GBq/m2 or standard marrow dose based on blood.