The purpose of this study was to assess the effect of dose reduction and the potential of noise reduction filters on image quality and the detection of liver lesions using MSCT. Twenty-nine patients with a total of 40 liver lesions underwent 16-slice CT (120 kV; 180 mAs). Virtual noise was added to CT raw datasets simulating effective mAs levels of 155, 130, 105, 80, 55, 30 and 10 mAs. All datasets were post-processed with an edge-preserving noise-reduction filter (ANR-3D), yielding a total of 15 datasets per patient. Ten radiologists performed independent evaluations of image quality, the presence of liver lesions and diagnostic confidence. Quantitative noise and contrast-to-noise ratios (CNR) were obtained. Superior image quality (P < 0.02), reduction of image noise (P < 0.001) and the increase of lesion-to-liver CNR (P < 0.001) were observed in images processed with the ANR-3D filter. Sensitivity for lesion detection remained unchanged down to 105 mAs (CTDI(w) 6.6 mGy) without filter and 80 mAs (CTDI(w) 5.1 mGy) with ANR-3D. Confidence was rated significantly higher for datasets reconstructed with ANR-3D. The use of a noise-reducing, but edge-preserving filter (ANR-3D) is a promising option to reduce further the radiation dose in liver CT.