Aiming　at　the　problem　that　it　is　difficult　to　acquire　the　feature　information　of　magnetic　memory　signal　for　low　speed　and　heavy　load　gear　under　fault　state,　a　new　method　based　on　the　intrinsic　time-scale　decomposition　(ITD)　is　proposed　to　achieve　the　extraction　of　magnetic　memory　signal　feature.　Firstly,　the　original　magnetic　memory　signals　are　decomposed　into　several　proper　rotation　components　(PRC)　and　a　monotonous　tendency　item　with　ITD　method.　Then　the　first　four　order　PRCs　are　reconstructed　to　eliminate　the　large　cycle　composition　in　magnetic　memory　signal　and　magnetic　noise.　Finally,　the　magnetic　signal　strength　of　each　gear　tooth　root　is　extracted　using　cycle　average　and　local　statistic　method.　Experiment　results　show　that　the　proposed　method　is　suitable　for　accurately　picking　up　and　judging　the　effective　compositions　of　the　signal,　it　can　effectively　extract　signal　feature　and　has　important　application　value　in　potential　fault　diagnosis　of　low　speed　and　heavy　load　gear.