Radioactive 90 Y-selective internal radiation (SIR) sphere therapy is increasingly used for the treatment of nonresectable hepatocellular carcinoma (HCC). However, the maximum delivered dose is limited by severe injury to the nontarget tissue, including liver parenchyma. Our study aimed to implement radiobiologic models for both tumor control probability (TCP) and normaltissue complication probability (NTCP) to describe more effectively local response and the liver toxicity rate, respectively. Methods: Patients with documented HCC, adequate bone marrow parameters, and regular hepatic and pulmonary function were eligible for the study. Patients who had pulmonary shunt greater than 20% of 99m Tc-labeled macroaggregated albumin or any uncorrectable delivery to the gastrointestinal tract, reverse blood flow out of the liver, or complete portal vein thrombosis were excluded. Patients received a planned activity of the 90 Y-SIR spheres, determined using the empiric body surface area method. The dose distribution was determined using posttreatment (3-dimensional) activity distribution and Monte Carlo dose voxel kernel calculations, and the mean doses to healthy liver and tumor were calculated for each patient. Response was defined according to Response Evaluation Criteria in Solid Tumors (RECIST) and recommendations of the European Association for the Study of the Liver (EASL). Criteria were used to assess possible liver toxicities. The parameters of TCP and NTCP models were established by direct maximization of the likelihood. Results: Seventy-three patients were treated. With an average dose of 110 Gy to the tumor, complete or partial response was observed in 74% and 55% of patients according to the EASL guideline and RECIST, respectively, and the predicted TCPs were 73% and 55%, respectively. With a median liver dose of 36 Gy (range, 6-78 Gy), the $grade 2 (G2), $grade 3 (G3), and $grade 4 (G4) liver toxicities were observed in 32% (23/73), 21% (15/73), and 11% (8/73) of patients, respectively. The parameters describing the $G2 liver toxicity data using the NTCP model were a tolerance dose of the whole organ leading to a 50% complication probability of 52 Gy (95% confidence interval, 44-61 Gy) and a slope of NTCP versus dose of 0.28 (95% confidence interval, 0.18-0.60), assuming n 5 1. Conclusion: The radiobiologic approach, based on patient-specific dosimetry, could improve the 90 Y-microsphere therapeutic approach of HCC, maintaining an acceptable liver toxicity.