In this study Hidden Markov models are employed for the automatic classification of pitch accents in German utterances. A simplified version of the tone sequence model (i.e. a linguistic theory of pitch accents) is applied. In this approach only two of the nuclear tones (rise, fall) are used. They are represented by continuous density Hidden Markov models. The classifier works on the sequence of feature vectors defined by N-tuples of succeeding fundamental frequency estimates.