This study demonstrated that dual energy x-ray absorptiometry can be used to distinguish between normal union and atrophic nonunion, with high sensitivity and high negative predictive value, by 8 weeks after surgery in a canine model. Eighteen adult mixed-breed dogs were divided into two equal groups: normal union and atrophic nonunion. In the normal union group, a 5 mm mid-diaphyseal transverse ostectomy was performed in the right tibia, and the bone was stabilized with a unilateral external fixator. In the atrophic nonunion group, a 5 mm mid-diaphyseal ostectomy was performed; the distal 1.5 cm of the bone ends, including the periosteum, were frozen twice to -20 degrees C using liquid nitrogen and thawed slowly twice; and the bone was stabilized with a unilateral external fixator. The members of the research team were blinded to the group assignments until after all dogs were killed and all data were acquired. Radiography and dual energy x-ray absorptiometry of both tibiae were performed at weeks 1, 2, 3, 4, 6, 8, 10, 12, 14, and 16 after surgery. All dogs were killed at 16 weeks, and the torsional stiffness and maximum torque of both tibiae of five dogs in each group were determined. Mechanically, the tibiae in the normal union group had significantly higher maximum torque (43-fold higher) and torsional stiffness (86-fold higher) than the tibiae in the atrophic nonunion group. The sensitivity and negative predictive value of dual energy x-ray absorptiometry for predicting nonunion were 100% by 8 weeks after surgery. The specificity and positive predictive value reached 78 and 82%, respectively, by 16 weeks. Radiographic scores were significantly higher for the union group than for the nonunion group beginning at 2 weeks after surgery. The earliest time after surgery that radiography classified an ostectomy as a nonunion was significantly later (4.6 +/- 1.2 weeks) than for dual energy x-ray absorptiometry (2.6 +/- 1.4 weeks). Before these results can be extrapolated to human applications, further animal studies must be performed to evaluate clinically or experimentally induced fractures, or both, rather than the well defined ostectomies that were performed in this study.