This paper describes a non-invasive method for classifying the motions of each finger using tactile features for the purpose of controlling a motorized prosthetic hand for the forearm. Tactile sensors were developed using polyvinylidene difluoride (PVDF) to detect tactile features on forearm surface caused by motion of fingers. The two tactile sensors were placed on the extensor digitorum and flexor digitorum superificialis on the forearm, and feature values were extracted from the measured signals. Using the obtained features as input, the motions were classified using machine learning. In this paper, comparisons were made using five different classifiers. Experimental results showed that the linear support vector machine obtained an average classification rate of more than 86% by using only two PVDF tactile sensors. It was also found that thirteen trials were required per type of motion.