The load rating for the bridge in operation is an important problem to help the project be safe. However, one of the difficulties encountered when solving this problem do not know exactly the actual parameters of the bridge. In this paper, load ratings for a steel box girder bridge built around 15 years are carried out. This paper studies the application of an artificial neural network to update the boundary conditions for the finite element model from the field vibration measurement results. Then, based on the actual boundary conditions of the bridge, load ratings for the bridge are carried out. The results of the paper show that the load rating of the bridge will be more accurate for the updated model.