Wi-Fi　is　widely　used　for　network　communication　in　our　daily　life　here　and　there.　In　addition　to　communication,　it　is　also　the　reference　information　that　is　widely　used　in　indoor　positioning,　such　as　localization　based　on　RSSI　(Radio　Signal　Strength　Indicator)　fingerprint　database　and　range　based　localization.　However,　the　broadcasting　signals　from　wireless　access　point　suffers　from　the　complex　indoor　environment,　such　as　multi-path　fading　and　non-homogeneous　environment.　And　the　received　signals＇　strength　at　the　mobile　terminal　is　instable,　which　must　influence　the　positioning　precision　in　indoor　localization.　Although　these　factors　are　existed,　localization　based　on　RSSI　fingerprint　database　are　still　used　widely　indoors.　There　are　2　stages　in　the　RSSI　fingerprint　database　localization,　offline　fingerprint　database　building　and　online　positioning.　And　the　quality　of　both　fingerprint　database　and　real　time　RSSI　values　influence　the　precision　of　localization.　However,　the　RSSI　based　indoor　localization　algorithms　existed　are　almost　conducted　on　the　original　fingerprint　database　which　is　not　optimized　Some　abnormal　RSSI　values　in　fingerprint　database　must　destroy　the　distribution　of　the　RSSI　road　map　and　result　in　low　precision.　In　order　to　decrease　the　influence　of　original　data　containing　bizarre　RSSI　possibly,　filtering　technology　should　be　applied　for　the　database.　Considering　the　geographic　correlation　among　RSSI　values　in　road　map,　the　neighbor　mean　filter　(NMF)　is　introduced　to　optimize　the　fingerprint　database　in　this　paper.　The　idea　of　NMF　is　simply　to　replace　each　pixel　value　in　an　image　with　the　mean　value　of　its　neighbors,　excluding　itself.　Thus,　it　has　the　advantage　for　filtering　the　bizarre　RSSI　values　in　fingerprint　database.　In　the　paper,　the　introduction　of　NMF　algorithm　is　described　firstly,　then,　the　bizarre　value　of　RSSI　fingerprint　database　is　analyzed　and　the　optimization　on　fingerprint　database　with　NMF　is　narrated,　especially,　the　method　of　determining　threshold　for　optimization.　Second,　the　indoor　localization　with　optimized　fingerprint　database　is　put　forward,　which　includes　the　architecture　for　RSSI　fingerprint　database　localization,　cosine　calculation　of　RSSI　vectors　and　their　similarity　judgement,　and　the　localization　based　on　optimized　fingerprint　database.　Through　these　measures,　the　abnormal　RSSI　values　found　by　NMF　method　and　replaced　with　mean　RSSI　values　could　make　the　fingerprint　database　more　accurate　for　the　indoor　positioning.　In　the　end,　with　Java　programming　language,　the　positioning　system　based　on　optimized　fingerprint　database　is　developed　in　Eclipse　environment.　Then　both　original　RSSI　fingerprint　database　and　the　database　filtered　are　tested　and　experimented　in　the　positioning　system,　and　the　result　shows　that　the　optimized　RSSI　fingerprint　database　can　increase　the　accurate　of　indoor　localization.