Objective: To investigate effects of long-term exercise therapy for people with knee osteoarthritis (OA)on radiographic disease severity and cartilage integrity (primary outcomes)as well as severity of bone marrow lesions (BMLs), synovitis and/or effusion (secondary outcomes). Methods: We sourced randomized controlled trials in people with clinical and/or radiographic OA investigating the effect of land-based exercise therapy of > 6 months on radiographic disease severity and magnetic resonance imaging outcomes of cartilage integrity (morphology or composition)as well as BML, synovitis and/or effusion severity, when compared to no exercise. Two independent reviewers extracted data and assessed risk of bias. Random-effects meta-analysis was used to pool standardised mean differences (SMD)(95% confidence intervals (CI))or odds ratios (OR)(95% CI)and estimate heterogeneity (I 2 , %). Quality of the pooled body of evidence was rated implementing the GRADE approach. Studies unsuitable for meta-analysis were summarized in a best-evidence synthesis. Results: Meta-analysis showed moderate quality evidence of no treatment effect on tibiofemoral radiographic disease severity ((SMD)95% (CI): 0.06 (−0.07, 0.20), I 2 = 0%)and low-quality evidence of no effect on tibiofemoral cartilage morphology (SMD (95%): 0.06 (−0.20, 0.36), I 2 = 0%). Low quality evidence revealed no treatment effect on the odds of change in synovitis ((OR)(95% CI): 0.90 (0.51,1.60), I 2 = 0%)and effusion ((OR (95% CI): 0.88 (0.64, 1.20), I 2 = 0%), but greater odds of tibiofemoral BMLs worsening (OR (95% CI): 1.90 (1.11, 3.26), I 2 = 0%). In best-evidence synthesis, limited evidence was found for changes in patellar cartilage composition following exercise in women with mild knee OA compared to no exercise, but not for tibiofemoral cartilage. Conclusion: Long-term exercise therapy did not change tibiofemoral radiographic disease severity, cartilage morphology or synovitis/effusion, but may slightly increase the likelihood for increased BML severity. Overall, meta-analysis findings were limited in directness and precision and restricted to relatively imprecise effect estimates in people who were obese on average. Limited evidence suggested some protective effects on patellar cartilage composition.