Proteasomes are enzymes in eukaryotic cells which cut proteins marked for degradation into fragments. In mammals some of these fragments are used by the immune system to detect proteins of foreign, e.g. viral, origin. Hence reproducing, predicting and possibly understanding the cleaving patterns of proteasomes is an interesting theoretical problem and its solution would be beneficial for vaccine design. The equations connecting cut probabilities, fragment frequencies and so-called cut strengths are derived. A simple model for the time course of protein digestion is used to explain the problem of fragment competition and the possible deviation of in vitro fragment frequencies from those that can be expected in vivo. A family of neural network proteasome models for the reproduction and prediction of cleavage patterns is described in detail together with the webtool PAProC. The first model is based on the experimentally observed cleavage pattern, an intermediate model on the distinction between weak and strong cuts, and the most elaborate model uses quantitative data, i.e., fragment frequencies.