Abstract: This paper proposes a data-driven intonation modeling method. With the principal component analysis (PCA) technique, the concept of eigenintonation is presented. The distribution of emotional states in the eigenintonation space is then studied and the corresponding emotional intonations in Mandarin are given out. In order to avoid the influences caused by the unwanted factors, the affective corpora for the purpose of evaluation are designed. With the corpora, the related experiments have been performed. The experimental results show that the eigenintonation model has a quite good ability of expressing emotions, and all intonation samples in our corpora can be well recovered with the eigenintonations.