The　update　of　rules　is　the　key　to　success　for　rule-based　network　intrusion　detection　system　because　of　the　endless　appearance　of　new　attacks.　To　efficiently　extract　classification　rules　from　the　vast　network　traffic　data,　this　paper　gives　a　new　approach　based　on　Particle　Swarm　Optimization　(PSO)　and　introduces　a　new　coding　scheme　called　＂indexical　coding＂　in　accord　with　the　feature　of　the　network　traffic　data.　PSO　is　a　novel　optimization　technique　and　has　been　shown　high　performance　in　numeric　problems,　but　few　researches　have　been　reported　in　rule　learning　for　IDS　that　requires　a　high　level　representation　of　the　individual,　this　paper　makes　a　study　and　demonstrates　the　performance　on　the　1999　KDD　cup　data.　The　results　show　the　feasibility　and　effectiveness　of　it.