The major purpose of this research was to determine whether kinesthetic/proprioceptive perceptions of the earth-fixed vertical axis are more accurate than perceptions of intrinsic axes. In one experiment, accuracy of alignment of the forearm to earth-fixed vertical and head- and trunk-longitudinal axes by seven blindfolded subjects was compared in four tasks: (1) Earth-Arm--arm (humerus) orientation was manipulated by the experimenter; subjects aligned the forearm parallel to the vertical axis, which was also aligned with the head and trunk longitudinal axis; (2) Head--head, trunk, and upper-limb orientations were manipulated by the experimenter, subjects aligned the forearm parallel to the longitudinal axis of the head using only elbow flexion/extension and shoulder internal/external rotation; (3) Trunk--same as (2), except that subjects aligned the forearm parallel to the trunk-longitudinal axis; (4) Earth--same as (2), except that subjects aligned the forearm parallel to the earth-fixed vertical. Head, trunk, and gravitational axes were never parallel in tasks 2, 3, and 4 so that subjects could not simultaneously match their forearm to all three axes. The results showed that the errors for alignment of the forearm with the earth-fixed vertical were lower than for the trunk- and head-longitudinal axes. Furthermore, errors in the Earth condition were less dependent on alterations of the head and trunk orientation than in the Head and Trunk conditions. These data strongly suggest that the earth-fixed vertical is used as one axis for the kinesthetic sensory coordinate system that specifies upper-limb orientation at the perceptual level. We also examined the effects of varying gravitational torques at the elbow and shoulder on the accuracy of forearm alignment to earth-fixed axes. Adding a 450 g load to the forearm to increase gravitational torques when the forearm is not vertical did not improve the accuracy of forearm alignment with the vertical. Furthermore, adding small, variably sized loads (between which the subjects could not distinguish at the perceptual level) to the forearm just proximal to the wrist produced similar errors in aligning the forearm with the vertical and horizontal. Forearm-positioning errors were not correlated with the size of the load, as would be expected if gravitational torques affected forearm-position sense. We conclude that gravitational torques exerted about the shoulder and elbow do not make significant contributions to sensing forearm-orientation relative to earth-fixed axes when the upper-limb segments are not constrained by external supports.