Accurate discrimination of changes in physiological response has been noted as an important element in learning to control that response. Using a magnitude production procedure involving forearm isometric contractions, it was found that the ability to discriminate muscle tension varied across headache groups, with the no-headache control group (r = .76) being most accurate, followed by the tension group (r = .68), the mixed tension and migraine group (r = .60), and finally the migraine group (r = .51). The most important result, however, was that muscle discrimination ability significantly predicted clinical outcome from treatment by relaxation training for tension headache subjects (r = .50) but did not predict outcome for migraine or mixed headache subjects.