OBJECTIVE: There is growing understanding of the importance of bone in OA. Our aim was to determine the relationship between 3D MRI bone shape and total knee replacement (TKR). METHODS: A nested case-control study within the Osteoarthritis Initiative cohort identified case knees with confirmed TKR for OA and controls that were matched using propensity scores. Active appearance modelling quantification of the bone shape of all knee bones identified vectors between knees having or not having OA. Vectors were scaled such that -1 and +1 represented the mean non-OA and mean OA shapes. RESULTS: Compared to controls (n = 310), TKR cases (n = 310) had a more positive mean baseline 3D bone shape vector, indicating more advanced structural OA, for the femur [mean 0.98 vs -0.11; difference (95% CI) 1.10 (0.88, 1.31)], tibia [mean 0.86 vs -0.07; difference (95% CI) 0.94 (0.72, 1.16)] and patella [mean 0.95 vs 0.03; difference (95% CI) 0.92 (0.65, 1.20)]. Odds ratios (95% CI) for TKR per normalized unit of 3D bone shape vector for the femur, tibia and patella were: 1.85 (1.59, 2.16), 1.64 (1.42, 1.89) and 1.36 (1.22, 1.50), respectively, all P < 0.001. After including Kellgren-Lawrence grade in a multivariable analysis, only the femur 3D shape vector remained significantly associated with TKR [odds ratio 1.24 (1.02, 1.51)]. CONCLUSION: 3D bone shape was associated with the endpoint of this study, TKR, with femoral shape being most associated. This study contributes to the validation of quantitative MRI bone biomarkers for OA structure-modification trials.