Internet traffic classification plays an important role in the different aspects of network management such as balancing bandwidth, assuring Quality of Service (QoS) and implementing billing mechanisms. Nowadays detection and classification of P2P traffic is an important concern of I and network administrators. The P2P applications are growing widely and they act greedy, consuming bandwidth as much as they can.The P2P host can easily share its content for other peers so it makes additional traffic in its LAN. Sharing and distributing the contents, makes peers near to each other however causes viruses and threats disseminate rapidly. It is important to restrict the P2P traffic for security policies. The P2P protocols use obfuscation methods to hide their traffic from filters and pass firewalls. These protocols use encryption, random ports and make detection more difficult. The encryption makes content of payload inaccessible and causes the content-based approaches will be ineffective. New methods which do not employ deep packet iection, are suitable for detecting encrypted protocols. They use information of headers in third and fourth layers which are feasible but not encrypted. Packet size and inter-arrival time between packets of flow, are the two important information of those headers. The obfuscation can also happen in the header information by means of changing and padding more bytes to packets. The combination of approaches is more effective, since Relying on only one method can not handle payload and header obfuscations. The combination of methods which uses information of payload and header of packets, is called hybrid method. In this thesis the hybrid method is used to classify encrypted protocols. In this classification, the content-based method is combined with the method which uses the header information. By using these two methods obfuscated protocols are detected with high accuracy. In this approach the part of content-based classifier measures the randomness of data in payloads of packets. For measuring the randomness, two techniques are implemented and the new one which is proposed in this thesis is more accurate than the other. After comparing the two new techniques the header information is added to the features so the hybrid classification is implemented Key Words:Internet traffic classification, statistical features, hybrid method, machine learning