Due to the complicated relationship between a muscle activation level and isometric joint torques, predicting torques under isometric conditions have been an obstacle to neuromuscular control of robots in the field of human robot interaction (HRI). This study describes the isometric torque estimation of the metacarpophalangeal (MCP) joint of a human hand based on the Hill muscle model. Surface electromyography (sEMG) for muscle activation and ultrasound imaging technique for muscle length changes of the first dorsal interosseous (FDI) muscle were employed under isometric condition. The result showed that the prediction is more dependent on sEMG than length change and has a high correlation value (r = 0.9604) with the measured torque. This result indicates the possibility of a torque prediction method based on the Hill muscle model for natural human robot interaction.