In　order　to　deal　with　the　problem　that　measurement　of　biochemical　oxygen　demand　(BOD)　in　wastewater　treatment　process　(WWTP)　is　difficult　to　achieve,　a　soft　sensing　algorithm　for　effluent　BOD　concentration　prediction　based　on　maximum　information　coefficient　(MIC)　and　radial　basis　function　neural　network　(RBFNN)　is　proposed.　Firstly,　the　MIC　is　employed　to　filter　the　input　variables　that　have　close　correlation　with　the　effluent　BOD　concentration.　Secondly,　an　improved　K-means　algorithm　is　used　to　initialize　the　center　and　width　of　the　RBFNN,　and　the　Levenberg-Marquardt　(LM)　algorithm　is　used　to　train　the　weight　of　the　network.　Finally,　the　benchmark　datasets　and　the　real　data　of　the　WWTP　are　used　for　experiments,　the　results　indicate　that　the　network　has　a　good　prediction　on　the　effluent　BOD　concentration.