In　the　last　decade,　the　high-speed　rail　(HSR)　has　undergone　rapid　development　and　is　playing　a　more　and　more　important　role　in　the　transportation　system　of　China.　However,　the　currently　adopted　maintenance　policy　of　HSR　is　still　mainly　usage-based　preventive　maintenance,　which　is　quite　conservative　and　incurs　tremendous　annual　maintenance　costs.　Thus,　it　is　necessary　to　conduct　predictive　maintenance　so　as　to　save　maintenance　cost　as　well　as　ensure　the　reliability　of　HSR,　which　requires　for　predicting　the　remaining　useful　life　(RUL)　as　an　essential　step.　As　sensor　technology　and　the　5th　generation　wireless　technology　advance,　condition　monitoring　has　been　convenient　and　cost　efficient.　Based　on　the　collected　condition　information　data,　the　RUL　prediction　becomes　possible.　In　this　research,　we　develop　an　Elman　artificial　neural　network　for　the　purpose　of　predicting　the　RUL　of　HSR　bearings,　based　on　the　condition　monitoring　data.　To　fulfill　this　purpose,　we　firstly　propose　the　concepts　of　current　and　cumulative　state　characteristics　for　analyzing　the　state　monitoring　data　to　extract　and　filter　features　that　can　reflect　the　current　state　of　the　bearings.　Then,　we　build　the　Elman　artificial　neural　network,　evaluate　the　role　cumulative　state　characteristics　play　in　the　model　and　obtain　the　weights　and　thresholds　with　optimal　prediction　performance.　This　way,　the　network　structure　and　the　neuron　number　of　hidden　layers　are　optimized.　Experimentation　based　on　the　data　set　of　the　2012　IEEE　PHM　Data　Challenge　demonstrates　the　goodness　of　the　proposed　approach.　©　Springer　Nature　Singapore　Pte　Ltd　2019.