ABSTRACT In a forearm position matching task in the horizontal plane, when one (reference) arm is conditioned by contraction and length changes, subjects make systematic errors in the placement of their other, indicator arm. Here we describe experiments that demonstrate the importance not just of conditioning the reference arm, but of the indicator arm as well. Total errors from muscle conditioning represented up to a quarter of the angular range available to subjects. The sizes of the observed effects have led us to repeat other, previously reported experiments. In a matching task in the vertical plane, when muscles of both arms were conditioned identically, if the subject supported their arms themselves, or when the arms were loaded by the addition of weights, the loading did not introduce new position errors. To test the effect of exercise, subjects’ elbow flexors were exercised eccentrically or concentrically by asking them to lower or raise a set of weights using forearm muscles. The exercise produced 25% - 30% falls in maximum voluntary contraction strength of elbow flexors and this led to significant position matching errors. The directions and magnitudes of the errors were similar after the two forms of exercise and indicated that subjects perceived their exercised muscles to be longer than they actually were. To conclude, the new data from loading the arm is not consistent with the idea that the sense of effort, accompanying support of a load, provides positional information, in any simple way. Our current working hypothesis is that when muscles are active, position sense involves operation of a forward internal model. Loading the arm produces predictable changes in motor output and afferent feedback while changes after exercise are unpredictable. This difference leads to exercise-dependent errors.