The linear attenuation coefficient (mu in cm-1) of trabecular bone was modeled for different conditions of bone and marrow composition in order to assess their influence on computed tomography (CT) quantitation. A large relative change (10% of TBV at 15% TBV) of bone concentration resulted in small changes of mu: 2.3% at 60 keV, 3.4% at 44 keV, 5.2% at 29 keV. Relative changes of trabecular bone volume (TBV) on the order of 3% could be detected in vivo by CT were it not for errors of relocation and for compositional influences on accuracy. The mu (and density) depended critically not only on amounts of bone substance and marrow but on their compositions. Normal variation in the composition of bone substance produced an uncertainty in mu equivalent to 0.5 to 1% TBV. Increases of yellow marrow produced a decrease of mu which could be mistaken for a decrease of bone concentration. The biological variation (90% confidence limit) of marrow composition gives an uncertainty at 15% TBV of about 2.4% TBV at 60 keV, 1.7% at 44 keV, and 1.3% at 29 keV. These correspond to relative uncertainties of 16, 11, and 9% respectively. These factors help explain the large accuracy errors (30%) observed in all studies of trabecular bone where single-energy CT was used. Marrow composition also can affect precision of bone measurement. Systematic shifts of red and yellow marrow could mask biological changes such as those occurring with aging or treatment.