This paper describes a modeling technique of Thai tones for HMM-based speech synthesis. Tones are important prosodic features for tonal language including Thai because the phonetically same words but with different tones give different meanings. Although there have been several approaches to improving tone correctness of synthetic speech by considering tone types, another significant factor, stress, was not used explicitly for prosody modeling. We incorporate stress/unstress information into the framework of the HMM-based speech synthesis. Objective and subjective evaluation results show that the use of stress information improves the performance in Thai tone modeling.