The stiffness of activated muscles may stabilize a loaded joint by preventing perturbations from causing large displacements and injuring the joint. Here the elbow muscle recruitment patterns were compared with the forearm loaded vertically (a potentially unstable inverted pendulum configuration) and with horizontal loading. Eighteen healthy subjects were studied with the forearm vertical and supinated and the elbow flexed approximately 90 degrees. In the first experiment EMG electrodes recorded activity of biceps, triceps, and brachioradialis muscles for joint torques produced (a) by voluntarily exerting a horizontal force isometrically (b) by voluntarily flexing and extending the elbow while the forearm was loaded vertically with 135N. The relationship between the EMG and the torque generated was quantified by the linear regression slope and zero-torque intercept. In a second experiment a vertical load increasing linearly with time up to 300N was applied. In experiment 1 the EMG-torque relationships for biceps and triceps had an intercept about 10% of maximum voluntary effort greater with the vertical compared to the horizontal force, the inverse was found for Brachioradialis, but the EMG-torque slopes for both agonist and antagonistic muscles were not different. In experiment 2 there were 29 trials with minimal elbow displacement and all the three muscles activated on the order of 11% of maximum activation to stabilize the elbow; 19 trials had small elbow extension and 14 trials small flexion requiring altered muscle forces for equilibrium; 7 trials ended in large unstable displacement or early termination of the test. An analysis indicate that the observed levels of muscle activation would only provide stability if the muscles' short-range stiffness was at the high end of the published range, hence the elbow was marginally stable. The stability analysis also indicated that the small elbow extension increased stability and flexion decreased stability.