The estimation error due to variations in soft-tissue baseline in lumbar bone mineral content (BMC) measured by dual-photon absorptiometry (DPA) was calculated with a new method of automatic baseline subtraction. In water phantom measurements, the s.d. of the soft-tissue (ST) baseline matched closely (r = 0.98) to the random error, calculated using 44 keV and 100 keV count rates and the directly determined baseline variations. In 21 volunteers and in 70 patients with osteoporosis, the ST variations were larger than the expected random error, revealing a source of error related to the inhomogeneity of soft tissue. The estimation error in BMC caused by ST variations was 0.7% in healthy subjects (mean BMC 40.5 gHA) and 1.5% in patients (mean BMC = 26.4 gHA). These results indicate that ST-related errors are an important limit to the precision of lumbar DPA measurements.