Spatial　outliers　represent　spatial　objects　whose　non-spatial　attributes　are　significantly　different　from　their　spatial　neighborhoods＇.　Spatial　outlier　detection　can　provide　user　with　unexpected,　interesting　and　useful　spatial　patterns　for　further　analysis　and　has　received　a　lot　of　attention.　However,　many　existing　methods　for　spatial　outlier　mining　use　the　k-neighborhood　method　to　determine　spatial　neighborhood　which　depends　on　a　priori　parameter　k,　and　don＇t　consider　spatial　autocorrelation.　As　a　result,　it　usually　violates　the　true　situation.　So　we　propose　a　similarity　measurement　between　spatial　objects　that　based　on　Delaunay　Triangulation　(DT_SOF),　which　captures　spatial　correlation　and　spatial　neighborhood　from　the　dataset　itself.　Furthermore,　the　measure　takes　in　account　the　local　behavior　of　a　spatial　object　in　its　neighborhood　Finally,　experimentations　on　a　synthetic　and　a　real-world　ecological　geochemical　dataset　demonstrate　that　our　approach　can　effectively　detect　spatial　outliers　with　a　lower　disturbance　by　human　intervention.