There are occasional marked discordances in BMD T-scores at the lumbar spine (LS) and femoral neck (FN). We investigated whether such discordances could contribute independently to fracture prediction using FRAX. We studied 21,158 women, average age 63 years, from 10 prospective cohorts with baseline FRAX variables as well as FN and LS BMD. Incident fractures were collected by self-report and/or radiographic reports. Extended Poisson regression examined the relationship between differences in LS and FN T-scores (ΔLS-FN) and fracture risk, adjusted for age, time since baseline and other factors including FRAX 10-year probability for major osteoporotic fracture calculated using FN BMD. To examine the effect of an adjustment for ΔLS-FN on reclassification, women were separated into risk categories by their FRAX major fracture probability. High risk was classified using two approaches: being above the National Osteoporosis Guideline Group intervention threshold or, separately, being in the highest third of each cohort. The absolute ΔLS-FN was greater than 2 SD for 2.5% of women and between 1 and 2 SD for 21%. ΔLS-FN was associated with a significant risk of fracture adjusted for baseline FRAX (HR per SD change = 1.09; 95% CI = 1.04-1.15). In reclassification analyses, only 2.3-3.2% of the women moved to a higher or lower risk category when using FRAX with ΔLS-FN compared with FN-derived FRAX alone. Adjustment of estimated fracture risk for a large LS/FN discrepancy (>2SD) impacts to a large extent on only a relatively small number of individuals. More moderate (1-2SD) discordances in FN and LS T-scores have a small impact on FRAX probabilities. This might still improve clinical decision-making, particularly in women with probabilities close to an intervention threshold.