Massive　training　samples　are　usually　difficult　to　obtain　in　practice　for　remaining　useful　life　(RUL)　prediction　of　rolling　element　bearings　(REBs).　Building　simulated　data　sets　is　an　alternate　solution.　Some　effective　dynamic　models　have　been　established　for　instantaneous　vibration　behaviors　of　REBs.　However,　few　researches　have　been　done　on　the　modeling　of　their　long-term　degradation　processes.　Thus,　plenty　of　degradation　data　are　obtained　by　dynamic　modeling　in　this　paper,　and　a　novel　performance　degradation　dictionary　(PDD)　is　constructed　for　RUL　prediction　based　on　similarity.　Firstly,　the　degradation　process　is　divided　into　steady　stage,　defect　initiation,　defect　propagation　and　damage　stage.　Considering　the　coupling　excitation　of　time-varying　morphology　and　stiffness,　a　comprehensive　dynamic　model　is　established　to　simulate　the　fault　propagation.　Secondly,　a　PDD　which　serves　as　the　reference　sets　for　RUL　prediction　is　constructed　by　solving　the　response　of　the　model.　Hence,　the　similarity　method　can　be　enhanced　to　estimate　the　uncertainty　of　RUL.　Finally,　the　effectiveness　of　the　proposed　method　is　verified　by　experimental　degradation　data　under　different　working　conditions　and　fault　types.　(C)　2020　Elsevier　Ltd.　All　rights　reserved.