Objectives: To compare 2-year outcomes of total knee replacement (TKR) followed by non-surgical treatment to that of non-surgical treatment alone and outcomes of the same non-surgical treatment to that of written advice. Design: In two randomized trials, 200 (mean age 66) adults with moderate to severe knee osteoarthritis (OA), 100 eligible for TKR and 100 not eligible for TKR, were randomized to TKR followed by non-surgical treatment, non-surgical treatment alone, or written advice. Non-surgical treatment consisted of 12 weeks of supervised exercise, education, dietary advice, use of insoles, and pain medication. The primary outcome was the mean score of the Knee Injury and Osteoarthritis Outcome Score (KOOS) subscales, covering pain, symptoms, activities of daily living (ADL), and quality of life (QOL). Results: Patients randomized to TKR had greater improvements than patients randomized to non-surgical treatment alone (difference of 18.3 points (95% CI; 11.3 to 25.3)), who in turn improved more than patients randomized to written advice (difference of 7.0 points (95% CI; 0.4 to 13.5)). Among patients eligible for TKR, 16 (32%) from the non-surgical group underwent TKR during 2 years and among those initially ineligible, seven patients (14%) from the non-surgical group and ten (20%) from the written advice group underwent TKR. Conclusions: TKR followed by non-surgical treatment is more effective on pain and function than non-surgical treatment alone, which in turn is more effective than written advice. Two out of three patients with moderate to severe knee OA eligible for TKR delayed surgery for at least 2 years following non-surgical treatment. Trial registration: ClinicalTrials.gov numbers NCT01410409 and NCT01535001.