In computer tomography (CT) diagnostics, the measured Hounsfield units (HU) are used to characterize tissue and are in that respect compared to nominal HU values found in the radiological literature. Quality assurance (QA) phantoms are commercially available with a variety of tissue substitutes and materials to test the HU values in CT. It is however recognized from CT physics that the HU for a given material is energy dependent and may vary substantially between scanners. The aim of this study is to analyze the characteristics of a commonly used QA phantom, the Catphan 500/600 (The Phantom Laboratory, NY). Four CT phantoms were scanned on one CT scanner to examine possible interphantom variations in HU values. Secondly, one selected phantom was scanned at three kVp levels on eight different CT scanners. The interphantom variations in HU values were small, in the range 2-5 HU. The interscanner variations were however substantial, in the range 7-56 HU depending on energy and material. Varying the x-ray energy produced a shift in the measured HU of up to 79 HU on one scanner. Reference HU values for the eight sensitometric test materials in Catphan are provided for eight CT scanner models from four vendors. The reference HU values are provided for 80, 120 and 140 kVp. Our results suggest that scanner-independent threshold levels for HU should be used only with extreme caution. Tissue characterization can be used provided that a scanner-specific data set for normal and abnormal is determined.