1. We investigated the role of stretch reflexes in controlling two antagonist muscles acting at the interphalangeal joint in the normal human thumb. Reflex action was compared when either muscle contracted alone and during cocontraction. 2. The total torque of the flexor pollicis longus (FPL) and extensor pollicis longus (EPL) muscles was measured in response to an externally imposed extension of the interphalangeal joint. The initial joint angle and the amplitude of the extension were constant in all experiments, and the preload of the active muscle(s) was varied. Joint torque was measured at the peak of short-latency stretch reflex action during contraction of the FPL alone, contraction of the EPL alone, and during cocontraction. Incremental joint stiffness was calculated as the change in torque divided by the change in angle. 3. Incremental stiffness increased in proportion to the preload torque during single muscle contractions of either the FPL (lengthening disturbances) or the EPL (shortening disturbances). Thus stiffness was not regulated to a constant value in the face of varying loads for either single muscle stretch or release. 4. Incremental stiffness varied across the range of cocontraction levels while the net torque was maintained at approximately 0. Thus net torque alone did not determine the stiffness during cocontraction. 5. The contributions of each muscle to the net intrinsic torque during cocontraction were estimated by scaling the individual muscles' responses so that their sum gave the best fit (in a least-squares sense) to the cocontraction torque before reflex action. The solution is unique because the individual torques have opposite signs, but the stiffnesses add. This gave estimates of the initial torques of both muscles during cocontraction. 6. The contributions of the two muscles during cocontraction were used to estimate the active joint stiffness that would be expected if the two muscles were activated independently to the same levels as in the cocontraction trials. The stiffness measured at the peak of stretch reflex action during cocontraction trials differed from the sum of the stiffnesses of the two muscles when they were contracting alone. At low cocontraction levels, the measured stiffness was less than expected on the basis of summation of the action of the two muscles, whereas at high cocontraction levels, the measured stiffness was greater than expected. This demonstrates that there is nonlinear stretch reflex interaction. That is, reflex action for a pair of antagonists is not simply the linear sum of the reflex actions of the two muscles acting independently.(ABSTRACT TRUNCATED AT 400 WORDS)