In this paper we present a novel algorithm that, given a short snippet of an audio performance (piano music, for the time being), identifies the piece and the score position. Instead of using audio matching methods we propose a combination of a state-of-the-art music transcription algorithm and a new symbolic fingerprinting method. The resulting system is usable in both on-line and off-line scenarios and thus may be of use in many application areas. As the evaluation shows the system operates with only minimal lag and achieves high precision even with very short queries.