Abstract Background To evaluate whether the common risk factors and risk scores (FRAX, QFracture, and Garvan) can predict hip fracture in the oldest old (defined as people aged 80 and older) and to develop an oldest-old-specific 10-year hip fracture prediction risk algorithm. Methods Subjects aged 80 years and older without history of hip fracture were studied. For the derivation cohort (N = 251, mean age = 83), participants were enrolled with a median follow-up time of 8.9 years. For the validation cohort (N = 599, mean age = 85), outpatients were enrolled with a median follow-up of 2.6 years. A five-factor risk score (the Hong Kong Osteoporosis Study [HKOS] score) for incident hip fracture was derived and validated, and its predictive accuracy was evaluated and compared with other risk scores. Results In the derivation cohort, the C-statistics were .65, .61, .65, .76, and .78 for FRAX with bone mineral density (BMD), FRAX without BMD, QFracture, Garvan, and the HKOS score, respectively. The category-less net reclassification index and integrated discrimination improvement of the HKOS score showed a better reclassification of hip fracture than FRAX and QFracture (all p < .001) but not Garvan, while Garvan, but not HKOS score, showed a significant over-estimation in fracture risk (Hosmer–Lemeshow test p < .001). In the validation cohort, the HKOS score had a C-statistic of .81 and a considerable agreement between expected and observed fracture risk in calibration. Conclusion The HKOS score can predict 10-year incident hip fracture among the oldest old in Hong Kong. The score may be useful in identifying the oldest old patients at risk of hip fracture in both community-dwelling and hospital settings.