Audio Fingerprints (AFP's) are compact, content-based representations of audio signals used to measure distances among them. An AFP has to be small, fast computed and robust to signal degradations. In this paper an entropy based AFP is presented that performed very well when the signal was corrupted with lossy compression, scaling and even 1KHz Low-pass filtering in the experiments. The AFP is determined by computing the instantaneous amount of information of the audio signal in two-second frames with fifty percent overlapping, the resulting entropy signal is binary coded in order to compare different interpretations (e.g. live vs. studio recording) of the same song with good results. The AFP's robustness is compared with that of Haitsma-Kalker's Hash string based AFP with encouraging results.