This paper is a study on the development of a wearable sensor system for reading Braille. The study is intended to develop a compact tactile sensor system which uses a PVDF (Polyvinylidene Fluoride) film for the sensory receptor. The sensor is mounted onto a fingertip and moved over Braille manually to obtain the output. Since the unsteady movements yield variations in the waveforms, a robust recognition system is required. The performance of the recognition system is highly affected by the transformation method and the classification method. As the investigation of the transformation method, feature vectors of some different forms are extracted and effective forms are explored. In the classification, two methods based on MLPs (Multi-Layerd Perceptrons) are examined. The first method uses plural classifiers, each of which uses the different form of vectors. The next method uses MLPs in the form of dichotomizers to enhance the training quality. The performance of each method is evaluated using test signals. The obtained results show that the robust recognition is achieved appropriately.