Skilled human operators can adjust various conditions of manufacturing system and deal with unexpected troubles flexibly and appropriately. However, the operators can possibly fail the adjustment and dealing at times, because of fatigue, carelessness, bad physical condition, etc. The failure may cause some serious delay of the manufacturing processes and even physical danger to the operators. So, an intelligent system for avoiding the failure or warning its signs is desired to improve their working environment. To develop the system, a reliable recognition technique of operator's motion is indispensable. In this research, the recognition system of five kinds of motion in cylindrical grinding machine operation has been investigated by utilizing a small gyro/acceleration sensor and DP matching method. The similarity between the reference motion (standard pattern) and an input motion has been evaluated by the index of DP matching, or the pattern distance. The reference motions have been derived by averaging several typical motion data after time normalization. As the result, to recognize the objective five motions, first, the similarity indices by only angle data should be compared each other. Then, the similarity indices by only acceleration data should be compared, if necessary.