Gaze estimation is an indispensable technology for constructing systems that use information about attention and is utilized in the operation of welfare robots and equipment. Generally, it requires head-mounted devices or glasses type sensor, however there are problems of giving restraint or uncomfortable feeling to the user. If a webcam could be used to estimate the gazing area, it is possible to obtain a gaze information without wearing the equipment while maintaining a wide field of vision. Therefore, an algorithm for estimating the gazing area with a webcam using appearance-based method is proposed in this study. In order to estimate the gazing area, convolutional neural network is adopted, and an original dataset is created for training the model. The experimental results show that a weighted centroid method improves the accuracy of gaze estimation and reduce the error in the estimated distance.