Activity distribution between wrist movers during rhythmic flexion-extension of the wrist has been analysed in three different mechanical conditions. Wrist angular position and surface EMG from Extensor Carpi Radialis (ECR) and Flexor Carpi Radialis (FCR) were recorded. In the first condition (hand prone, flexion-extension in a vertical parasagittal plane) the hand passive equilibrium position was approximately 50 degrees in flexion. During hand oscillations FCR and ECR were alternatively recruited to move the hand symmetrically away from the equilibrium and de-recruited to allow conservative forces to restore the equilibrium. Switching between antagonists occurred at the centre of the oscillation (equilibrium crossing). In the second condition (hand semi-prone, flexion-extension in a horizontal transversal plane) the hand equilibrium was attained over an angle of about 26 degrees . When the hand was oscillated symmetrically around this equilibrium range, each muscle was recruited when the hand entered the equilibrium range and switching between antagonists therefore occurred in advance of the oscillation centre. Both vertical and horizontal oscillations were also performed all externally to the equilibrium position or range: in these cases only one muscle was recruited over the entire cycle, the EMG burst starting at the onset of the related movement. In the third condition (hand semi-prone, flexion-extension in a horizontal transversal plane) a frictional load added to the platform pivot expanded the equilibrium range to encompass the entire hand oscillation. Now concentric muscle contraction was needed throughout each phase of the movement and switching between antagonists occurred at the movement reversal, i.e. ~90 degrees in advance of the oscillation centre. The above descriptions held for oscillation frequencies from 0.2 Hz to 3.0 Hz, once the frequency-dependent effects of viscosity and inertia were accounted for. In all the three conditions, contractile forces started developing when an intrinsic or external resistance had to be overcome in order to continue the movement. To account for this control, a neural network is proposed that compares the afferent information about joint position with a position central command, thus detecting the position error caused by the forces that resist to movement. From the sign and amplitude of the error signal the network determines the direction (agonist vs antagonist) and the amount of motor activation.