Perceptual coders promise to deliver true transparent coding. The listener should not be able to hear any difference between original and reconstructed signal. The main elements of perceptual coders are (1) a filterbank to map the time domain input signal to a time- frequency domain representation, (i.e., the masked threshold) and (2) the quantization and coding step that should include proper noise shaping according to the psychoacoustic model. An overview of current and future audio coding systems is given. The completion of ISO/IEC 11172-3 (the MPEG-1 audio coding standard) marks the end of the work on the first generation of audio coding systems. While audio coding is on the way to widespread application, topics for the new coding systems include joint stereo coding of two or more channels, different analysis-by-synthesis systems, and improved psychoacoustic models.