A machine control device includes: first decoders for estimating, from brain signal information, which one of body movements a user performs or images, based on learning using pairs of movements performed by the user and brain signal information, the body movements going toward one of first to nth body postures; second decoders for estimating from the brain signal information, an correct rate on the body movement estimation, based on learning using pairs of correct rates of the estimation of body movements and the brain signal information; and electric prosthetic arm control section for controlling an electric prosthetic arm to change stepwise its posture between first to nth postures via at least one intermediate posture therebetween, the first to nth postures corresponding to the first to nth body postures. The first decoders perform the estimation only when the estimated correct rate exceeds a threshold. When the first decoders estimate that the body movements goes toward a body posture different from the current one, the electric prosthetic arm control section changes the posture of the machine by performing a part of substeps of change from one of the first to nth body postures toward the posture associated with the body posture toward which the estimated body movement goes. This configuration allows a user to control a brain-signal-based machine to perform a natural movement without a long-term training and much brain information.