1. Estimates of pennation in human muscles are usually obtained from cadavers. In this study, pennation of human brachialis was measured in vivo using sonography. Effects of static and dynamic changes in elbow angle and torque were investigated. 2. Pennation was measured in eight subjects using an 80 mm, 5 MHz, linear-array ultrasound transducer to generate sagittal images of the brachialis during maximal and submaximal isometric contractions at various elbow angles. It was shown that estimates of pennation were reproducible, representative of measurements made throughout the belly of the muscle and not distorted by compression of the muscle with the transducer or rotation of the muscle out of the plane of the transducer. 3. Mean resting pennation was 9.0 +/- 2.0 deg (S.D., range 6.5-12.9 deg). When the muscle was relaxed there was no effect of elbow angle on pennation. However, during a maximal isometric contraction (MVC), with the elbow flexed to 90 deg, pennation increased non-linearly with elbow torque to between 22 and 30 deg (mean 24.7 +/- 2.4 deg). The effect of increasing torque was small when the elbow was fully extended. The relationship between elbow angle, elbow torque and brachialis pennation suggests that the relaxed brachialis muscle is slack over much of its physiological range of lengths. 4. There was no hysteresis in the relationship between torque and pennation during slow isometric contractions (0.2 MVC s-1), and the relationship between elbow angle and pennation was similar during slow shortening and lengthening contractions. 5. Two consequences follow from these findings. Firstly, intramuscular mechanics are complex and simple planar models of muscles underestimate the increases in pennation which occur during muscle contraction. Second, spindle afferents from relaxed muscles may not encode joint angle over the full range of movement.