This paper demonstrates how a cognitive process that is usually simulated with a symbolic rule system can be modeled with a so-called PDP (Parallel Distributed Processing) network. Netzsprech, a model that learns to transcribe German text into phonemes (similar to NetTalk by Sejnowski & Rosenberg, 1986), serves as an example to show the advantages of the PDP approach. The rule-following beahvior of the model is compared directly with symbolic rule systems. As a result, it can be argued that networks of that kind will be a powerful complement for AI programs in the future.