Objective: This study investigated the effect of exercise therapy on inflammatory activity in synovitis and bone marrow lesions (BMLs) assessed by magnetic resonance imaging (MRI) in patients with knee OA. Methods: 60 patients with knee OA were randomized 1:1 to 12 weeks of supervised exercise therapy 3 times/week (ET) or a no-attention control group (CG). Synovitis and BMLs were assessed with static MRI with and without contrast and with dynamic contrast enhanced MRI (DCE-MRI). DCE-MRI data was quantified using pixel-by-pixel methodology based on analysis of signal intensity curves. Pain was assessed by the Knee Injury and Osteoarthritis Outcome Score (KOOS). Analyses of covariance were used assessing group differences in changes from baseline to week 12. Results: 33 patients adhered to the protocol and had valid MRI and KOOS data (ET, n = 16, CG, n = 17). Statistically significant and clinically relevant group difference in favour of ET was seen in KOOS pain change (−11.7 points, 95%CI: −20.1 to −3.4). There were statistically significant group differences in DCE-MRI assessed synovitis in the anterior synovium with unchanged inflammatory activity in the ET group compared to the CG. There were no group differences in BMLs and static MRI. Conclusion: Inflammatory activity was unchanged, and pain was reduced in patients with knee OA adhering to 12 weeks of exercise therapy compared to a no-attention control group. The reduction in pain was not explained by changes in inflammatory activity. Overall, the results suggest that exercise is not harmful in knee OA. ClinicalTrials.gov number: NCT01545258.