Atomic　clock　frequency　difference　prediction　is　the　key　step　in　atomic　clock　time　scale　calculation　and　atomic　clock　control.　Precise　prediction　algorithm　can　accurately　predict　the　future　operation　state　of　atomic　clock　which　can　be　used　to　improve　the　accuracy　of　atomic　time.　In　order　to　further　improve　the　prediction　accuracy　of　atomic　clock　frequency　difference,　a　genetic　wavelet　neural　network　(GAWNN)　atomic　clock　frequency　difference　prediction　algorithm　is　proposed　in　this　paper,　which　is　based　on　wavelet　neural　network　(WNN)　atomic　clock　frequency　difference　prediction　algorithm.　The　genetic　algorithm　is　used　to　optimize　the　wavelet　neural　network　so　as　to　select　the　appropriate　number　of　hidden　layers　and　the　number　of　training　points　to　construct　the　atomic　clock　frequency　difference　prediction　model.　In　this　paper,　the　algorithm　is　validated　by　the　hydrogen　clock　and　cesium　clock　actual　frequency　difference　data　of　the　National　Institute　of　Metrology,　and　the　results　show　that　the　algorithm　improves　the　prediction　accuracy　of　hydrogen　clock　and　cesium　clock　frequency　difference　data.　©　2017　IEEE.