A general framework for probabilistic prognosis using maximum entropy approach, MRE, is proposed in this paper to include all available information and uncertainties for RUL prediction. Prognosis metrics are used for model comparison and performance evaluation. Several conclusions can be drawn based on the results in the current investigation: The proposed MRE updating approach results in more accurate and precise prediction compared with the classical Bayesian method. The classical Bayesian method is a special case of the proposed MRE approach and MRE approach is more flexible to include additional information for inference, which cannot be handled by the classical Bayesian method. The prognosis metrics can be successfully used for algorithm comparison and can give quantitative values in model (algorithm) performance evaluation. A robustness metric measuring the updating algorithmic sensitivity to prior uncertainty is proposed and applied to both Bayesian and MRE updating approaches. The application examples show that MRE exhibits more robustness against the uncertainty introduced by parameter distribution priors in the sense of prognosis performance. It is important to realize when to apply these metrics to arrive at meaningful interpretations. For instance, use of the convergence metric makes sense only when the algorithm predictions converge (get better) with time.