Background: Gait modifications can reduce the knee adduction moment, a representation of knee loading. Reduced loading may help to slow progression of medial knee osteoarthritis. We aimed to investigate the response of patients with medial knee osteoarthritis to direct feedback on the knee adduction moment as a method for modifying the gait pattern, before and after training with specific gait modifications. Methods: Forty patients with medial knee osteoarthritis underwent 3D gait analysis on an instrumented-treadmill, while receiving real-time feedback on the peak knee adduction moment. Patients were trained with three different modifications; toe-in, wider steps and medial thrust gait. The response to real-time feedback on the knee adduction moment was measured before and after training. To evaluate the short term retention effect, we measured the changes without feedback. We also evaluated the effects on the knee flexion moment and at the hip and ankle joints. Findings: With direct feedback on the knee adduction moment, patients were initially unable to reduce the knee adduction moment. After training with specific modifications, peak knee adduction moment was reduced by 14% in response to direct feedback. Without feedback a 9% reduction in peak knee adduction moment was maintained. Hip moments were not increased with modified gait, but small increases in ankle adduction moment and knee flexion moment were observed. Interpretation: Real-time biofeedback directly on the knee adduction moment is a promising option for encouraging gait modifications to reduce knee loading, however only when combined with specific instructions on how to modify the gait.