The aim of this study was to assess the accuracy of virtual noncontrast (VNC) images of the liver in a phantom and patients using dual-source photon-counting detector computed tomography (PCD-CT). An anthropomorphic abdominal phantom with a liver insert containing liver parenchyma (1.4 mgI/mL) and 19 liver lesions (iodine content 0-5 mgI/mL) was imaged on a clinical dual-source PCD-CT (tube voltage 120 kV) and in the dual-energy mode on a dual-source energy-integrating detector (EID) CT (tube voltage combinations, 80/Sn150 kV, 90/Sn150 kV, and 100/Sn150 kV). Rings of fat-equivalent material were added to the phantom to emulate 3 sizes (small, medium, large). Each setup was imaged at 3 different radiation doses (volume CT dose index: 5, 10, and 15 mGy). Virtual noncontrast images were reconstructed and CT attenuation was measured in each lesion and liver parenchyma. The absolute error of CT attenuation (VNCerror) was calculated using the phantom specifications as reference. In addition, 15 patients with hypodense liver lesions who were clinically scanned on PCD-CT were retrospectively included. Attenuation values in lesions and liver parenchyma in VNC images reconstructed from portal venous phase CT were compared with true noncontrast images. Statistical analysis included analysis of variance with post hoc t tests and generalized linear models to assess the impact of various variables (dose, patient size, base material, iodine content, and scanner/scan mode) on quantification accuracy. In the phantom, the overall mean VNCerror for PCD-CT was 4.1 ± 3.9 HU. The overall mean VNCerror for EID-CT was 7.5 ± 5, 6.3 ± 4.7, and 6.7 ± 4.8 HU for 80/Sn150 kV, 90/Sn150 kV, and 100/Sn150 kV, respectively, with the VNCerror of EID-CT being significantly higher at all tube voltage settings (P < 0.001), even after adjusting for dose, size, iodine content of the lesion, and attenuation of base material. For PCD-CT, a smaller phantom size was associated with higher quantification accuracy (P = 0.007-0.046), whereas radiation dose did not impact accuracy (P > 0.126). For EID-CT, but not for PCD-CT, VNCerror increased with lesion iodine content (P < 0.001). In patients, there was no difference in attenuation measured on true noncontrast and VNC images (P = 0.093), with a mean VNCerror of 3.7 ± 2.2 HU. Photon-counting detector CT allows for the reconstruction of VNC images of the liver both in a phantom and in patients with accurate attenuation values, being independent of dose, attenuation of base material, and liver iodine content.