Brain　activation　detection　is　an　important　problem　in　fMRI　data　analysis.　In　this　paper　we　propose　a　data-driven　activation　detection　method　called　neighborhood　one-class　SVM　(NOC-SVM).　By　incorporating　the　idea　of　neighborhood　consistency　into　one-class　SVM,　the　method　classifies　a　voxel　as　an　activated　or　non-activated　voxel　by　its　neighbor　weighted　distance　to　a　hyperplane　in　a　high-dimensional　kernel　space.　On　two　synthetic　datasets　under　different　SNRs,　the　proposed　method　almost　has　lower　error　rate　than　K-means　clustering　and　fuzzy　K-means　clustering.　On　a　real　fMRI　dataset,　all　the　three　algorithms　can　detect　similar　activated　regions.　Furthermore,　the　NOC-SVM　is　more　stable　than　random　algorithms,　such　as　K-means　clustering　and　fuzzy　K-means　clustering.　These　results　show　that　the　proposed　NOC-SVM　is　a　new　effective　method　for　activation　detections　in　fMRI　data.