Non-contact　geomagnetic　anomaly　detection,　as　one　of　passive　non-destructive　testing　(NDT)　techniques,　can　be　used　to　locate　pipeline　defects,　while　its　accuracy　is　affected　by　random　noise　and　detection　orientation.　In　order　to　extract　effective　geomagnetic　anomaly　signals　of　pipeline　defects,　a　method　based　on　empirical　mode　decomposition　(EMD)　and　magnetic　gradient　tensor　was　studied.　In　order　to　filter　random　noise,　EMD　was　performed　to　self-adaptively　decompose　magnetic　field　signals　into　a　series　of　intrinsic　mode　functions　(IMFs),　and　then　Hurst　exponent　was　implemented　to　exclude　false　modes;　The　calculation　method　of　magnetic　gradient　tensor　modulus　(MGTM)　was　proposed　to　obtain　precise　defect　locations　according　to　tensor　symmetry;　Subsequently,　the　remote　pipeline　defect　model　was　built　based　on　the　magnetic　dipole　theory,　and　the　relationship　between　detection　orientation　and　MGTM　was　discussed.　The　experimental　results　showed　that　the　proposed　method　could　realize　high　precision　and　reliable　non-contact　geomagnetic　localization　of　pipeline　defects.