In this study, it is desired to estimate the torque held in an exoskeleton via electromyography signals. Several concepts were used during the study, such as: analog electronics for filtering and rectification of electromyography signals acquired; dynamical systems for modeling and estimation of the exoskeleton torque performed; neural network, which subsequent to its training and testing in different situations, was able to interpret the magnitude of the signal and bring it to the torque performed in the exoskeleton.