A study was conducted to observe the relationship between the architecture and function of the muscle-tendon complex in human movements. Using real time ultrasonography, we were able to observe clearly and noninvasively the movement of the fascicle and aponeurosis in human muscle in vivo, and measure directly the changes in pennation angle and fascicle length during muscle contraction. During dorsal and plantar flexion without additional load, the movement of tendinous tissue in the tibialis anterior muscle (TA) appeared to synchronize with the displacement of the ankle joint. On the other hand, when the ankle joint was fixed and the TA contracted "statically or isometrically", the ultrasonic echo from the deep aponeurosis in the TA was observed to move proximately, indicating that the elastic component (i.e. mainly tendinous tissue) was stretched significantly by the muscle contraction force. When the knee extensors contracted "isometrically", the fascicle length decreased at every joint angles and its magnitude was greater (30%) when the knee was closer to full extension than at flexed positions (5%). During "isokinetic" knee extension, the fascicle shortening velocity changed significantly due to the joint angle, i.e. the highest shortening velocity was observed at a knee joint angle of 60 degree. This indicates that the assumption of a constant muscle shortening velocity in isokinetic testing is not valid. In vertical jumping, the fascicle length of the gastrocnemius medialis muscle (GM) decreased by 27% in the first half of the push-off phase, when the whole length of the muscle-tendon complex (MTC) of the GM remained constant. On the other hand, before take-off the GM muscle contracted isometrically while the MTC length decreased. This means that in the first half of jumping, the elastic component is lengthened by the contraction force of the muscle fibers (elastic energy stored), and that in the following take-off phase MTC shortening is causked by the elastic energy stored previously when the muscle fibers contract isometrically. The present results clearly show that the architecture of actively contracting muscle fibers differs considerably from that which occurs when movement is passively induced. Therefore, the use of cadaver data in studies of the architecture and modeling of muscle functions would result in inaccurate, and in some cases even erroneous results.