This paper presents a novel method to control a robot arm based on the human skeletal information. First, the depth data is obtained by using a Kinect device. Skeletal tracking technology is used to deal with the depth data to match the various parts of the body and the 3D coordinates of each joint of the body are established. Then, the 3D coordinates are proposed to calculate the rotation angle of the human waist, shoulder, elbow and wrist joints. These four joints of the body are one-to-one correspondence with the four joints of 4-DOF robot arm. Finally, a transformation is conducted to map the calculation results into control commands, and the control commands are sent to the controller of robot arm by serial ports to achieve body feeling interaction. The experimental results show that the method can effectively control the robot arm to rotate and grab objects.