The implementation of directional pad functionalities on a glove-based optical fiber sensor by monitoring the thumb posture is reported. Multimode fiber bending transducers were attached to the glove to measure the flexion/extension and adduction/abduction movements of thumb joints. Then, the optical signals corresponding to each transducer were correlated to the relative finger position during the manipulation of a directional pad by using artificial neural networks. The estimation of thumb stationary location over the pad yielded a maximum absolute error of 4.3 mm, whereas the dynamic evaluation of finger trajectories resulted in errors lower than 5.5 mm. Furthermore, it was demonstrated that the glove sensor is able to identify the correct direction with 96.67% accuracy when applied as an eight-direction digital controller.