AIM: Transarterial radioembolization (TARE) is, by all standards, a radiation therapy. As such, according to Euratom Directive 2013/59, it should be optimized by a thorough treatment plan based on the distinct evaluation of absorbed dose to the lesions and to the non-tumoural liver (two-compartment dosimetry). Since the dosimetric prediction with (99m)Tc albumin macro-aggregates (MAA) of non-tumoural liver is much more accurate than the same prediction on lesions, treatment planning should focus on non-tumoural liver rather than on lesion dosimetry. The aim of this study was to determine a safety limit through the analysis of pre-treatment dosimetry with (99m)Tc-MAA single photon emission computed tomography (SPECT/CT), in order to deliver the maximum tolerable absorbed dose to non-tumoural liver. METHODS: Data from intermediate/advanced hepato-cellular carcinoma (HCC) patients treated with (90)Y glass microspheres were collected in this single-arm retrospective study. Injection was always lobar, even in case of bilobar disease, to avoid treating the whole liver in a single session. A three-level definition of liver decompensation (LD) was introduced, considering toxicity only in cases of liver decompensation requiring medical action (LD type C, LDC). We report LDC rates, receiver operating characteristic (ROC) analysis between LDC and NO LDC absorbed dose distributions, normal tissue complication probability (NTCP) curves and uni- and multivariate analysis of risk factors associated with toxicity. RESULTS: A 6-month timeline was defined as necessary to capture all treatment-related toxicity events. Previous transarterial chemoembolization (TACE), presence or extension of portal vein tumoural thrombosis (PVTT) and tumour pattern (nodular versus infiltrative) were not associated with tolerance to TARE. On the contrary, at the multivariate analysis, the absorbed dose averaged over the whole non-tumoural liver (including the non-injected lobe) was a prognostic indicator correlated with liver decompensation (odds ratio = 4.24). Basal bilirubin textgreater 1.1 mg/dL was a second even more significant risk factor (odds ratio = 6.35). NTCP analysis stratified with this bilirubin cut-off determined a 150 Gy/90 Gy for bilirubin textgreater/textless 1.1 mg/dL. These results are valid for a (90)Y glass microsphere administration 4 days after the reference time. CONCLUSION: Given the low predictive accuracy of (99m)Tc-MAA on lesion absorbed dose reported by several authors, an optimized TARE with (90)Y glass microspheres with lobar injection 4 days after reference time should aim at an absorbed dose averaged over the whole non-tumoural liver of 50 Gy/90 Gy for basal bilirubin higher/lower than 1.1 mg/dL, respectively.