In the past decade, the Internet has been widely used in everyday life. Different types of mobile broadband applications are created and require an increasing amount of network resources. However, Internet service providers must maximize the use of these limited resources to provide users with different levels of quality of service. The first step toward traffic engineering is to perform traffic classification. In this work, we propose a classification method for identifying mobile applications executed at an early stage. We first collected traffic traces from a WiFi access point and then developed a hidden Markov model based on the packet size and transmission direction of the first 20 packets. In a series of our experiments, we evaluated the number of hidden states through 10-fold cross validation, and classified six types of mobile applications. The accuracy of the proposed method achieved 99.17%. In addition, we set specific threshold values for different application models and identified 91.33% of flows in the testing data set, which comprised unknown traffic flows. These experimental results demonstrate that our proposed method is effective for classifying Internet flows as well as unknown traffic in a real network.