<p>Previous studies on the computational principle for solving the movement selection problem for the human arm have primarily focused on hand trajectories associated with the two-joint movements of the shoulder and elbow joints. Further, only a few computational models, that consider the musculoskeletal system, have been investigated. From this perspective, a minimum muscle-stress-change model was evaluated for the fingertip trajectories and arm postures during three-joint movements in the horizontal plane, including wrist joint rotation. A musculoskeletal model of a three-joint arm with eight muscles was used to perform the optimization calculations that determine the optimal arm movements. Results show that the computational model can reproduce the measured fingertip trajectories and arm postures to an equal or greater extent compared with the minimum angular-jerk model and the minimum torque-change model. Furthermore, the errors of the minimum muscle-stress-change model remained small for different values of joint viscosity, physiological cross-sectional areas, and moment arms, resulting in a small dependency of these parameters. In contrast, the minimum torque-change model resulted in considerable errors under low-viscosity conditions. Consequently, the minimum muscle-stress-change model has emerged as a promising candidate for elucidating the computational principle.</p>