Background: Knee osteoarthritis patients co-contract in knee-related muscle pairs during walking. The determinants of this co-contraction remain insufficiently clear. Methods: A heterogeneous group of 14 patients was measured before and one year after knee arthroplasty, and compared to 12 healthy peers and 15 young subjects, measured once. Participants walked on a treadmill at several imposed speeds. Bilateral activity of six muscles was registered electromyographically, and co-contraction time was calculated as percentage of stride cycle time. Local dynamic stability and variability of sagittal plane knee movements were determined. The surgeon's assessment of alignment was used. Pre-operatively, multivariate regressions on co-contraction time were used to identify determinants of co-contraction. Post-operatively it was assessed if predictor variables had changed in the same direction as co-contraction time. Findings: Patients co-contracted longer than controls, but post-operatively, differences with the healthy peers were no longer significant. Varus alignment predicted co-contraction time. No patient had post-operative varus alignment. The patients' unaffected legs were more unstable, and instability predicted co-contraction time in both legs. Post-operatively, stability normalised. Longer unaffected side co-contraction time was associated with reduced affected side kinematic variability. Post-operatively, kinematic variability had further decreased. Interpretations: Varus alignment and instability are determinants of co-contraction. The benefits of co-contraction in varus alignment require further study. Co-contraction probably increases local dynamic stability, which does not necessarily decrease the risk of falling. Unaffected side co-contraction contributed to decreasing affected side variability, but other mechanisms than co-contraction may also have played a role in decreasing variability. © 2011 Elsevier Ltd. All rights reserved.