Physical exercise induces spatially heterogeneous bone changes in the proximal femur. Recent advances have enabled 3D dual-energy X-ray Absorptiometry (DXA)-based finite element (FE) models to estimate bone strength. However, its ability to detect exercise-induced BMD and strength changes is unclear. The aim of this study was to quantify the repeatability of vBMD and femoral neck strength obtained from 3D-DXA images and determine the changes due an exercise intervention. The DXA scans included pairs of same-day repeated scans from ten healthy females and pre- and post-exercise intervention scans of 26 males. FE models with element-by-element correspondence were generated by morphing a template mesh to each bone. BMD and femoral strength under single-leg-stance and sideways fall loading configurations were obtained for both groups and compared. In the repeated images, the total hip vBMD difference was 0.5 ± 2.5%. Element-by-element BMD differences reached 30 ± 50%. The strength difference in single-leg stance was 2.8 ± 13% and in sideways fall was 4.5% ± 19%. In the exercise group, strength changes were 6 ± 19% under single-leg stance and 1 ± 18% under sideways fall. vBMD parameters were weakly correlated to strength (R2 < 0.31). The exercise group had a mean bone accrual exceeding repeatability values in the femoral head and cortical regions. The case with the highest vBMD change (6.4%) caused 18% and −7% strength changes under single-leg stance and sideways fall. 3D-DXA technology can assess the effect of exercise interventions in large cohorts but its validity in individual cases should be interpreted with caution.