Abstract: As a type of generalization of fuzzy sets, neutrosophic sets are helpful to describe all kinds of uncertain, inconsistent and discontinuous information. Since the classification result of the three-way decision on a single-valued neutrosophic set is relatively unitary, a three-way decision model on an interval-valued neutrosophic set is constructed. Firstly, based on the definition of a single-valued neutrosophic set, the concept of an interval-valued neutrosophic set is given, and a new three-way decision model is proposed by combining Bayesian decision theory and Manhattan distance. Then, under the given similarity, the conditional probability is calculated to obtain the expected loss in the information system of the interval-valued neutrosophic set, and the three-way decision model of the interval-valued neutrosophic set is established. Finally, the application of new model is illustrated by an example of medical diagnosis, and the effects of parameter variations on decision results are discussed.