OBJECTIVES: To compare the responsiveness of the generic Short Form 36 (SF-36) and the raw and Rasch-based scores of the condition-specific Oxford Knee Score (OKS) in patients undergoing total knee replacement (TKR) METHODS: Adult patients undergoing TKR in a hospital of Singapore between 2001 and 2006 completed the SF-36 and OKS at baseline and at 6 and 24 months postoperatively. OKS data were fitted to the Rasch partial credit model using the Winsteps program. Responsiveness was assessed using effect size (ES), standardised response mean (SRM), and relative validity (RV). RESULTS: A total of 702 patients who had complete data at baseline and two follow-ups were included in the analysis. After removing items regarding limping and kneeling, the remaining OKS items fit the Rasch model. Bodily pain (BP) and Physical functioning (PF) were more responsive than the other SF-36 domains. In addition, the OKS raw scores (raw-OKS) and Rasch-based modified OKS (Rasch-OKS) were consistently more responsive than all eight SF-36 domains. At the 6-month follow-up, Rasch-OKS had the largest ES whereas raw-OKS had the largest SRM (2.7 and 1.9, respectively). When compared to raw-OKS, the RV of Rasch-OKS, BP, and PF were 1.5, 2.0, 2.8, respectively. Similar order was observed at the 24-month follow-up. CONCLUSIONS: The OKS is more responsive than the SF-36 in patients undergoing total knee replacement. The raw and Rasch-based scores of OKS have comparable responsiveness. Different responsiveness indices may give different results.