During rapid movements, tendons can act like springs, temporarily storing work done by muscles and then releasing it to power body movements. For some activities, such as frog jumping, energy is released from tendon much more rapidly than it is stored, thus amplifying muscle power output. The period during which energy is loaded into a tendon by muscle work may be aided by a catch mechanism that restricts motion, but theoretical studies indicate that power can be amplified in a muscle-tendon load system even in the absence of a catch. To explore the limits of power amplification with and without a catch, we studied the bullfrog plantaris muscle-tendon during in vitro contractions. A novel servomotor controller allowed us to measure muscle-tendon unit (MTU) mechanical behavior during contractions against a variety of simulated inertial-gravitational loads, ranging from zero to 1× the peak isometric force of the muscle. Power output of the MTU system was load dependent and power amplification occurred only at intermediate loads, reaching ∼1.3× the peak isotonic power output of the muscle. With a simulated anatomical catch mechanism in place, the highest power amplification occurred at the lowest loads, with a maximum amplification of more than 4× peak isotonic muscle power. At higher loads, the benefits of a catch for MTU performance diminished sharply, suggesting that power amplification >2.5× may come at the expense of net mechanical work delivered to the load.