To overcome the drawbacks of existing methods for traffic classification (by ports, Deep Packet Inspection, statistical classification) a new system was developed, in which the data are collected and classified directly by clients installed on machines belonging to volunteers. Our approach combines the information obtained from the system sockets, the HTTP content types, and the data transmitted through network interfaces. It allows to group packets into flows and associate them with particular applications or types of service. This paper presents the design of our system, the implementation, the testing phase and the obtained results. The performed threat assessment highlights potential security issues and proposes solutions in order to mitigate the risks. Furthermore, it proves that the system is feasible in terms of uptime and resource usage, assesses its performance and proposes future enhancements. We released the system under The GNU General Public License v3.0 and published as a SourceForge project called Volunteer-Based System for Research on the Internet.