New generations of humanoid robots are becoming more faithful copies of the human body. From the human point of view, besides the human’s look, the interest of humanoid robot can be in the more natural communication and interaction with human. Humanoid is also well adapted to environment dedicated to human, home or company. With morphology close to human one, it can be expected that the humanoid can replace the human in its task, but also that it can acquire human skills. Our objective is to generate anthropometric motion of humanoid robot inspired by human motion in everyday human activities. This can be done by imitation process, but in this context, each task has to be learned independently. Thus our approach consists in the search of the criterion used by human to achieve a task since the joint motion for achieving the same task is not unique. We will consider motions of the upper-body of the robots in which each arm has 7 degrees of freedom. Criteria that be explored are the minimization of the joint displacement, minimization of the kinetic energy and the proximity to ergonomic configuration. A study at the kinematic level based on generalized pseudo inverse of the Jacobian matrix is implemented. The joint motions obtained by minimization of the criterion are compared to the human motion adapted to the kinematic of the humanoid. The choice of the criterion optimized by the human is the one that corresponds to the smallest difference on joint configuration during all motion. Four different types of motion are considered