The current state-of-the-art in pattern and character classification still reveals many unsolved problems, e.g., a robust classifier with respect to noise or other distortions in the character images is one of them. It is desirable that a classifier is easy to train and on the other hand very robust to any possible errors in the character images. Furthermore, the classification procedure should be real-time capable. In this technical report we introduce a new classifier that is based on trellis diagrams. It basically works like a Viterbi decoder known from communication systems. The fundamentals of the training and classification procedure are discussed in detail. In addition, we show the performance of the classifier on data with and without additive noise of different levels.