Feature　selection,　also　called　attribute　selection,　is　the　process　of　selecting　a　subset　of　features　from　a　relatively　large　dataset.　Feature　selection　is　efficient　in　removing　redundant　and　noisy　features　and　the　selected　features　are　effective　to　describe　the　whole　dataset.　Since　there　are　many　features　in　intrusion　detection　data,　which　is　large　in　quantity,　feature　selection　plays　an　important　role　in　intrusion　detection.　Within　these　features,　many　are　numerical　features.　The　traditional　feature　selection　methods,　such　as　the　correlation　analysis,　information　gain　(IG),　support　vector　machine　(SVM)　and　rough　set,　must　discretize　numerical　features　when　dealing　with　mixed　features.　The　discretization　process　usually　consumes　much　time　and　sometimes　will　even　result　in　the　loss　of　important　information,　decreasing　the　classification　accuracy.　Aiming　at　solving　these　problems,　the　neighborhood　rough　set　reduction　model　is　employed　in　this　paper,　which　can　process　the　numerical　features　directly　without　discretization.　Then　the　particle　fitness　function　in　particle　swarm　optimization　(PSO)　algorithm　is　built　based　on　that　model.　Finally,　a　novel　feature　selection　algorithm　based　on　particle　swarm　optimization　and　neighborhood　rough　set　reduction　model　is　proposed.　Experimental　results　prove　that　the　new　algorithm　improves　classification　ability　with　fewer　features　selected.