Prosthetic hands help upper limb amputees and people who were born without hands. Currently, these prostheses are rather rudimentary and do not provide adequate sensing capabilities compared to a human hand. People use their natural hands to perceive complex tactile phenomena such as shear and torsion using thousands of mechanoreceptors in their fingertips. The capability to detect torsional loads at the fingertips is a notable gap in prosthetic hand sensation. Flexible tactile sensors are a promising new technology that would be ideal for prosthetic hands since they allow for stretching and movement like human skin without damage to the sensor. Therefore, the purpose of this study is to determine whether a flexible magnetic sensor array combined with an artificial neural network (ANN) can detect and classify torsion. The flexible magnetic sensor is designed as a 3×3 array of magnets embedded in a stretchable elastomer which are situated atop a corresponding array of Hall effect sensors. Torques applied to the soft magnetic skin caused displacement of the magnetic fields that were perceived by the nine Hall effect sensors. In this study, ten different values of torque were applied to the flexible magnetic sensor array using a robotic arm to ensure consistency. Data were used to train an ANN to classify the applied torques. The ANN was trained ten times and could predict the applied torque with an average training classification accuracy of 97.48% ± 0.33%. Given the results of this study, this novel sensor design could enable more refined sensations of touch for people who use prosthetic hands.