Most measures of femoral neck strength derived using dual‐energy X‐ray absorptiometry or computed tomography (CT) assume the femoral neck is a cylinder with a single cortical thickness. We hypothesized that these simplifications introduce errors in estimating strength and that detailed analyses will identify new parameters that more accurately predict femoral neck strength. High‐resolution CT data were used to evaluate 457 cross‐sectional slices along the femoral neck of 12 postmortem specimens. Cortical morphology was measured in each cross‐section. The distribution of cortical thicknesses was evaluated to determine whether the mean or median better estimated central tendency. Finite‐element models were used to calculate the stresses in each cross‐section resulting from the peak hip joint forces created during a sideways fall. The relationship between cortical morphology and peak bone stress along the femoral neck was analyzed using multivariate regression analysis. In all cross‐sections, cortical thicknesses were non‐normally distributed and skewed toward smaller thicknesses (p < 0.0001). The central tendency of cortical thickness was best estimated by the median, not the mean. Stress increased as the median cortical thickness decreased along the femoral neck. The median, not mean, cortical thickness combined with anterior‐posterior diameter best predicted peak bone stress generated during a sideways fall (R2 = 0.66, p < 0.001). Heterogeneity in the structure of the femoral neck determines the diversity of its strength. The median cortical thickness best predicted peak femoral neck stress and is likely to be a relevant predictor of femoral neck fragility.