In this paper a methodology for bearing RUL estimation is proposed which is composed of five main steps: Data Collection, HI (Health indicator) Computation, HI Transformation, HMM (Hidden Markov models) Training, RUL (Remaining Useful Life) Estimation. Starting from a qualitative degradation model found in literature, an optimal HMM structure was identified by establishing a link between the bearing degradation phenomenon and the HMM. By doing so, the HMM model size was reduced. The obtained results show good prediction capabilities and a strong connection to the ongoing degradation phenomenon ensuring their correct interpretation.