During the last years, new technologies approaches have helped to develop realistic robotic hands for prosthetic use. Even so, the strategies to control them (input signals, prediction algorithms) are still limiting a complete match between the robotic hand and the real hand movements and behaviors. On this thesis, two different input signals (FMG and sEMG) were evaluated. From this analysis characteristic properties from each kind of signal were obtained, related with wrist and hand movements. In this way two different learning methods were implemented for the first time on robotic hand research. The goal of these two methods was to combine both kind of input signals, supported by the feature analysis previously done, in order to improve the movements prediction performance. The methods’ performance were compared with the separate input signals methods, so the improvement could be measured. Both mixing methods presented better results than the single input signal ones. These results along with other considerations defined, could lead to a robotic hand performance improvement from different perspectives