Abstract: In order to solve the problem of heavy workload and low accuracy in the manual monitoring of grassland grazing cattle, this paper proposed a grassland cattle movement behavior recognition method based on a binary decision-tree classification model. The P-R curve was constructed by selecting the X-axis, Y-axis, Z-axis variance, root mean square, average, and three-axis overall signal vector magnitude（SVM） and signal magnitude area（SMA） of the three-axis acceleration sensor data in the grassland cattle neck. The grouping method of the optimal behavior category and the optimal threshold corresponding to each statistical feature quantity were obtained by P-R curve.The information gain was used as the selection criterion to construct a binary decision-tree classification model, and this model was used to classify and recognize the four types of movement behaviors of prairie cattle: lying, ruminating, feeding and walking, which were compared with the K-means clustering algorithm.The results showed that the K-means clustering algorithm could only recognize lying behaviors, but it was difficult to distinguish three other kinds of behaviors: ruminating, feeding and walking.However, the binary decision-tree classification model could effectively identify lying and walking behaviors from ruminating and feeding behaviors. The accuracy rates and the precision rates reached 0.760 or more. The results indicated that the binary decision-tree classification model could complete the grassland cattle behavior classification more effectively than the commonly used K-means algorithm, and the accuracy rate was higher.