In human subjects, stiffness of the relaxed elbow was measured by three methods, using a forearm manipulandum coupled to a.d.c. torque motor. Elbow stiffness calculated from frequency response characteristics increased as the driving amplitude decreased. Step displacements of the forearm produced restoring torques linearly related to the displacement. The stiffness was very similar to that calculated from natural frequencies at amplitudes above 0.1 rad. Thirdly, elbow stiffness was estimated from brief test pulses, 120 ms in duration, by mathematically simulating the torque-displacement functions. Stiffness values in the limited linear range (under +/- 0.1 rad) were higher than in the linear range of the first two methods. A major component of elbow stiffness appears to decay within 1 s. The coefficients of viscosity determined from the simulation were, however, very similar to those calculated from the frequency response. Test pulse simulation was then used to determine joint impedance for different, actively maintained elbow angles. Joint stiffness and viscosity increased with progressive elbow flexion.