The recognition of emotions from speech is a challenging issue. In this paper, two Hidden Markov Model-based vocal emotion classifiers are trained and evaluated by an emotional mandarin speech corpus based on Mel-Frequency Cepstral Coefficient features. Up to 6 basic emotion models including angry, fear, happy, sad, neutral and surprise are built under different parameters and the influence of parameter set is investigated. A statistical comparison of the two emotion recognition methods are discussed as well. The overall results reveal that the GMM classifier outperforms HMM classifier taking both computation complexity and recognition rate into consideration with the highest recognition rate of 72.34%.