The　clock　difference　prediction　of　the　atomic　clock　is　one　of　the　parts　in　constructing　the　atomic　time　scale　and　controlling　the　time　scale,　so　it　is　very　important　to　explore　high-precision　clock　difference　prediction.　As　a　feedback　neural　network　that　can　dynamically　remember,　Elman　neural　network　is　very　suitable　for　time　series　prediction.　This　paper　proposes　an　algorithm　of　clock　difference　prediction　based　on　Elman　neural　network,　and　uses　the　relevant　data　of　the　hydrogen　clocks　and　cesium　clocks　from　National　Institute　of　Metrology　(China)　to　conduct　experiments　in　order　to　research　the　effectiveness　of　the　algorithm.　The　experimental　results　show　that　the　prediction　algorithm　which　is　proposed　in　this　paper　improves　the　prediction　accuracy　of　clock　difference　compared　with　the　prediction　algorithm　of　linear　regression,　the　prediction　algorithm　of　support　vector　machine　and　the　least　square　support　vector　machine.　©　2020　The　Institution　of　Engineering　and　Technology.