A total of 626 U.S. male Marine Corps recruits underwent anthropometric measurements and dual-energy X-ray absorptiometry (DXA) scans of the femoral midshaft and the distal third of the tibia prior to a 12 week physical training program. Conventionally obtained frontal plane DXA scan data were used to measure the bone mineral density (BMD) as well as to derive the cross-sectional area, moment of inertia, section modulus, and bone width in the femur, tibia, and fibula. During training, 23 recruits (3.7%) presented with a total of 27 radiologically confirmed stress fractures in various locations in the lower extremity. After excluding 16 cases of shin splints, periostitis, and other stress reactions that did not meet fracture definition criteria, we compared anthropometric and bone structural geometry measurements between fracture cases and the remaining 587 normals. There was no significant difference in age (p = 0.8), femur length (p = 0.2), pelvic width (p = 0.08), and knee width at the femoral condyles (p = 0.06), but fracture cases were shorter (p = 0.01), lighter (p = 0.0006), and smaller in most anthropometric girth dimensions (p < 0.04). Fracture case bone cross-sectional areas (p < 0.001), moments of inertia (p < 0.001), section moduli (p < 0.001), and widths (p < 0.001) as well as BMD (p < 0.03) were significantly smaller in the tibia and femur. After correcting for body weight differences, the tibia cross-sectional area (p = 0.03), section modulus (p = 0.05), and width (p = 0.03) remained significantly smaller in fracture subjects. We conclude that both small body weight and small diaphyseal dimensions relative to body weight are factors predisposing to the development of stress fractures in this population. These results suggest that bone structural geometry measurements derived from DXA data may provide a simple noninvasive methodology for assessing the risk of stress fracture.