The combination of functional MR imaging and novel robotic tools may provide unique opportunities to probe the neural systems underlying motor control and learning. Here, we describe the design and validation of a MR-compatible, 1 degree-of-freedom pneumatic manipulandum along with experiments demonstrating its safety and efficacy. We first validated the robot's ability to apply computer-controlled loads about the wrist, demonstrating that it possesses sufficient bandwidth to simulate torsional spring-like loads during point-to-point flexion movements. Next, we verified the MR-compatibility of the device by imaging a head phantom during robot operation. We observed no systematic differences in two measures of MRI signal quality (signal/noise and field homogeneity) when the robot was introduced into the scanner environment. Likewise, measurements of joint angle and actuator pressure were not adversely affected by scanning. Finally, we verified device efficacy by scanning 20 healthy human subjects performing rapid wrist flexions against a wide range of spring-like loads. We observed a linear relationship between joint torque at peak movement extent and perturbation magnitude, thus demonstrating the robot's ability to simulate spring-like loads in situ. fMRI revealed task-related activation in regions known to contribute to the control of movement including the left primary sensorimotor cortex and right cerebellum.