A general multi-model Bayesian network for fatigue damage prognosis is proposed in this paper. Uncertainties introduced by model choice, mechanism modeling, model parameter, and response measures are systematically included and hierarchically managed using a two-level Bayesian network. Additional relevant information is used to update the network state using the trans-dimensional Markov chain Monte Carlo (MCMC) simulations in the general multi-model state space. To improve the simulation efficiency, a new algorithm is developed to construct the proposal distributions in the trans-dimensional MCMC simulation. The model probabilities, parameter densities, Bayes factors, and the prognosis averaging are readily calculated based on the simulation results. A fatigue damage prognosis example incorporating three fatigue crack models is presented for methodology demonstration. Experimental data are used to validate the effectiveness of the method.