Previous experimental studies of failure of rotator cuff repair have involved a single pull to ultimate load. Such an experimental design does not represent the cyclic loading conditions experienced in vivo. We created 1 cm x 2 cm rotator cuff defects in 16 cadaver shoulders, repaired each defect with transosseous simple sutures, and cyclically loaded the repairs by a servohydraulic materials test system actuator at physiologic rates and loads (rate of 33 mm/sec to a load of 180 N.). A progressive gap was noted in each specimen, for a 100% rate of failure of the repairs. A 5 mm gap developed at an average of 25 cycles, and a 10 mm gap developed at an average of 188 cycles. The central suture always failed first and by the largest magnitude. This study suggests that rotator cuff tears that are repaired with a "tension overload" of a portion of the muscle-tendon units will undergo gradual failure with physiologic cyclic loading until the normal resting lengths of the muscle-tendon units are restored. This "controlled failure" of the repairs may explain residual defects that have been demonstrated by ultrasonography and by arthrography in patients with "successful" rotator cuff repairs. Clinical implications are that: 1) rotator cuff tears should be repaired without tension if possible; and 2) transosseous bone tunnels should ideally extend distal to the weak metaphysical bone so that purchase is obtained into cortical bone for greater fixation strength.