Using robotic orthoses to rehabilitate hemiparetic patients presents many challenges, especially concerning the control of such devices. Although most studies use electromyography (EMG) as the control signal, mechanomyography (MMG) has been showing great potential. This paper demonstrates an MMG-based hand gesture classifier developed using common methods usually used for processing and classifying EMG signals. Two Multilayered Perceptrons (MLP) were implemented to classify five hand positions and the resting state. Results were mixed, as the system could differentiate between the resting position and the other gestures with 67% accuracy, but when tasked with identifying the exact positions, the accuracy was lower than 20%. Although the model presented mixed results, great potential was observed in using MMG signals in conjunction with artificial neural networks for orthoses control.