In this paper, a corpus-based fuzzy hidden Markov model (FHMM) is proposed to model the speech act in a spoken dialogue system. In the training procedure, 29 FHMMs are defined and trained, each representing one speech act in our approach. In the identification process, the Viterbi algorithm is used to find the top M candidate speech acts. Then Bayes decision criterion, which stores the relationship between the phrase and the speech act, is employed to choose the most probable speech act from the top M speech acts. In order to evaluate the proposed method, a spoken dialogue system for air travel information service is investigated. The experiments were carried out using a test database from 25 speakers (15 male and 10 female). There are 120 dialogues, which contains 725 sentences in the test database. The experimental results show that the correct response rate can achieve about 82.7% using the FHMM and the Bayes decision criterion.