To　predict　the　available　capacity　and　state　of　health　of　lithium　ion　batteries　by　using　Gaussian　process　regression,　one　of　the　crucial　tasks　is　to　choosing　the　covariance　function.　This　paper　proposes　a　method　which　can　fulfill　the　Gaussian　process　regression　by　using　most　proper　covariance　functions　or　the　optimal　combination.　First,　a　variety　of　typical　functions　are　tried　to　fit　the　battery　experimental　data　points　with　least　square　method,　which　can　give　us　a　valuable　interpretation　of　the　properties　of　the　data;　next,　the　three　functions　with　minimum　root　mean　square　errors　are　selected　to　guide　the　choosing　of　the　patterns　of　the　covariance　functions;　then　the　Gaussian　process　regression　is　applied　on　the　training　data　to　determine　the　ultra-parameters　included;　finally,　we　use　the　Gaussian　process　model　to　predict　the　latter　cycle　capacities　within　the　test　data.　Experiments　show　that　the　combination　of　selected　covariance　functions　is　effective　and　can　be　applied　on　predictions　with　different　batteries.　Also,　the　method　can　reduce　the　time　in　applying　Gaussian　process　regression　by　determination　of　the　covariance　function　quickly.　©　2018　IEEE.