Gesture-based applications widely range from direct manipulation interfaces to speaking aids for the deaf. The crucial point in recognizing gestures is that it requires great computational power to deal with spatio-temporal patterns. In this paper, a syntactic approach is proposed to provide a simple recognition algorithm. In order to verify the proposed method, we apply it to recognize 3D arm movements involved in the Taiwanese Sign Language. We extract prime patterns from the input patterns. The classification is then accomplished by deciding which one of possible arm movements can produce the sequence of primary patterns. Experiments were conducted to confirm the effectiveness of the method.