The natural control of robotic prosthetic hands with non-invasive techniques is still a challenge: myoelectric prostheses currently give some control capabilities; the application of pattern recognition techniques is promising and recently started to be applied in practice but still many questions are open in the field. In particular, the effects of clinical factors on movement classification accuracy and the capability to control myoelectric prosthetic hands are analyzed in very few studies. The effect of regularly using prostheses on movement classification accuracy has been previously studied, showing differences between users of myoelectric and cosmetic prostheses. In this paper we compare users of myoelectric and bodypowered prostheses and intact subjects. 36 machine-learning methods are applied on 6 amputees and 40 intact subjects performing 40 movements. Then, statistical analyses are performed in order to highlight significant differences between the groups of subjects. The statistical analyses do not show significant differences between the two groups of amputees, while significant differences are obtained between amputees and intact subjects. These results constitute new information in the field and suggest new interpretations to previous hypotheses, thus adding precious information towards natural control of robotic prosthetic hands.