In recent years, driving assistance systems based on the prediction of driving behavior are becoming important for safe driving. A driver typically drives a vehicle following the procedure of recognition, decision and operation. Because a driver mainly recognizes the outside world from visual information, the gaze information will reflect the driver’s behavior earlier than the information obtained from the vehicle. Therefore, we propose a method of predicting a driving behavior using the driver’s gaze information. This method tries to predict six behaviors: left turn, right turn, lane change from right to left, lane change from left to right, going straight at a traffic intersection and stopping for a red light. The proposed method consists of two phases, namely, learning phase and predicting phase. In the learning phase, the method extracts features from gaze information and constructs a SVM classifier. Then, the method extracts the features from gaze information during driving, and predict the driving behavior using the constructed classifier. We evaluated the method with the gaze information obtained on an open road, and we confirmed its effectiveness. [Note] This document is an informal handout distributed at an IEICE TC-PRMU workshop.