Topic tracking, which starts from a few sample stories and finds all subsequent stories that discuss the same topic, is a new challenge for the text categorization task and makes a significant contribution to the accessibility of information, such as archives of news, e-mails, and historical newspapers. Much previous research on topic tracking uses machine learning techniques. However, the small size of the training data, especially positive training stories, presents diffi- culties in training the parameters of the tracking system to produce optimal results. In this paper, we present a method for topic tracking using subject templates to select an optimal training set. The method was tested on the TV news which are the outputs of a speech recognizer, and the result shows the effectiveness of the method.