Aim/Introduction: Converting time-integrated activity to absorbed dose using voxel-S-values (VSV) is an efficient method for voxel-based dosimetry yet its limitation in heterogeneous media is recognized. This study aims to assess the differences among different VSV methods as compared to Monte Carlo (MC) methods for lung, liver, lung-liver interface and tumor absorbed dose based on Y-90 microsphere TOF PET/CT images. Materials and Methods: This study included four sets of Y-90 microsphere TOF PET/CT patient data available from the University of Michigan Deep Blue Data Repository. The PET matrix size was 200*200*122 with a voxel size of 4.07*4.07*3 mm. VSV (Gy/MBq-s) were generated by GATE v.8.0 with a matrix size of 21*21*21 and same voxel size as PET images. Liver, lungs and tumors were manually segmented from CT images and mapped to the corresponding PET images. Tumors could only be delineated on two patients while liver and lungs could be segmented from all patients. Time integrated activity maps were generated from PET images assuming only physical decay after the acquisition. Investigated VSV methods were liver kernel with density correction (LKD), constant liver kernel (LK), central voxel scaling kernel (CK), lung+liver kernels (LLK) and local deposition (LD). MC results were used as the gold standard. Mean absorbed doses were calculated for tumors, liver, lungs, lung-liver interface, i.e., liver_i and lung_i which are 1.5 cm slabs of liver and lungs extended from the interface, respectively. Results: LKD, LK, CK, LLK and LD methods were within 3% of MC for mean liver and tumor absorbed dose. Compared to MC, their average differences were 26%±4.6%, -47.6%±36.2%, -8.6%±34.4%, -8.5%±2.0% and 19.9%±3.6%, for mean lung absorbed dose, 53.2%±10.8%, -61.7%±2.7%, 24.9%±5.7%, -12.8%±3.2% and 46.4%±5.8%, for lung_i, and 4.4%±3.3%, -4.4±3.4%, 6.1%±1.7%, 19.2%±8.0% and -6.7%±4.8% for liver_i. Conclusion: Absorbed doses from five VSVs are similar to MC for liver and tumors which are located far from the lung-liver interface. LKD and LD overestimate lung absorbed dose whereas others underestimate it. For the interface, LKD and LK perform reasonably well on liver_i, while LLK performs better on lung_i as compared to others. VSV poses more errors in lung absorbed dose as compared to the liver, especially manifests in the interface. LLK is suggested to obtain the lung absorbed dose, and LKD or LK is suggested to obtain the liver or tumor absorbed dose. References: [1] Mikell et al. EJNMMI physics, 2015; 2(1): 16. [2] Götz et al. Phys. Med. Biol 2019; 64(24):245011.