In the effort to make robot-assisted upper limb passive movement training effective for neurologic injuries suffered from stroke and spinal cord injury (SCI), a new fuzzy adaptive closed-loop supervisory control method for passive joint movement training is proposed. Firstly, high-level supervisory controller for the desired passive range of motion (PROM) is designed based on the impaired limb’s joint motion recovery, and then low-level closed-loop position tracking controller is presented to drive the robot stably and smoothly to stretch the impaired limb to move along the predefined trajectory. The suggested strategy was applied to the four degrees of freedom (DOF) Whole Arm Manipulator (WAM) rehabilitation robot to evaluate its performance. Experimental results carried out on the 4-DOF WAM rehabilitation robot show the effectiveness and potentialities of the fuzzy adaptive passive movement control in clinical application.