We tested a dual-energy bone densitometer (LUNAR DPX) that uses a stable x-ray generator and a K-edge filter to achieve the two energy levels. A conventional scintillation detector in pulse-counting mode was used together with a gain stabilizer. The densitometer normally performs spine and femur scans in about 6 minutes and 3 minutes, respectively, with adequate spatial resolution (1.2 x 1.2 mm). Total body scans take either 10 minutes or 20 minutes. The long-term (6 months, n = 195) precision of repeat measurement on an 18-cm thick spine phantom was 0.6% at the medium speed. Precision error in vivo was about 0.6, 0.9 and 1.5% for spine scans (L2-L4) at slow, medium and fast speeds, while the error was 1.2 and 1.5 to 2.0%, respectively, for femur scans at slow and medium speed. The precision of total body bone density was 0.5% in vitro and in vivo. The response to increasing amounts of calcium hydroxyapatite was linear (r = 0.99). The densitometer accurately indicated (within 1%) the actual amount of hydroxyapatite after correction for physiological amounts of marrow fat. The measured area corresponded exactly (within 0.5%) to that of known annuli and to the radiographic area of spine phantoms. There was no significant effect of tissue thickness on mass, area, or areal density (BMD) between 10 and 24 cm of water. The BMD values for both spine and femur in vivo correlated highly (r = 0.98, SEE = .03 g/cm2) with those obtained using conventional 153Gd DPA. Similarly, total body BMD correlated highly (r = 0.96, SEE = .02 g/cm2) with DPA results.