Some audio files' formats contain metadata such as musical genre. This facilitates the usability of musical devices. However there are still many audio files' formats in which no metadata about the sound can be found. Our goal is to provide to users the same comfort in these conditions by computing the genres automatically. Various techniques have been proposed to determine a song's genre among N known genres. In this paper, we propose a new way of using Markov models as classifiers to perform genres classification. Experiments on 10 genres including 4 cameroonian genres showed an accuracy of 69.4%.