Epilepsy is the most common psychological disorders in humans. Patients suffering from epilepsy usually experience behavioral symptoms, such as involuntary movement and rage reaction. Conventional monitoring with EEG and video is unpleasant for patients. Use of the accelerometer and electroencephalogram to access the state of vigilance has been discussed recently. In addition to simply detecting the occurrence of a seizure event or not, we proposed a concept of motor detection method which detects the subtle change of motor states in a kindling-induced seizure by using the accelerometer's signals in this study. The Wistar rats with temporal lobe epilepsy were used in our experiments, and the accelerometers were set on the rat's head. The features of 3-axis accelerometer signals were extracted, and the support vector machine (SVM) was applied to determine the motor states. The results show that our proposed on-line detection method can detect the changes of motor responses in a kindling-induced seizure with 78% accuracy in average. It has the advantage of easy measurement for real-world applicability.