The biomechanical mechanism of tibial diaphyseal stress fractures was studied prospectively in a group of 286 Israeli recruits. Before training each recruit had roentgenograms taken of his tibiae. Measurements of total tibial and cortical widths in the anteroposterior (AP) and mediolateral planes were made on these roentgenograms at two levels: at the point of the narrowest tibial width on AP roentgenograms (Level 1) and at the point of the narrowest width on lateral roentgenograms (Level 2). The tibial cross section was idealized as an eccentric ellipse within an ellipse, and on the basis of measurements taken from the roentgenograms, the cross-sectional area (compression strength), area moments of inertia about AP and mediolateral axes of bending (bending strength), and the area polar moment of inertia (torsional strength) were calculated for each cross section. During the course of 14 weeks of training, 20% of the recruits sustained tibial diaphyseal stress fractures, all of which were along the medial cortex. Using stepwise logistic regression analysis the tibia's bending strength along an AP axis of bedding at Level 2 was found to be the most significant factor determining whether or not a recruit would develop a tibial stress fracture.