Wastewater　treatment　process　(WWTP)　is　a　highly　nonlinear　dynamic　process.　It　is　difficult　for　modeling　key　parameters　of　WWTP.　In　order　to　measure　the　parameters,　a　new　recurrent　neural　network　(RNN)　with　novel　topology　is　proposed　in　this　paper.　The　proposed　RNN　is　a　class　of　locally　recurrent　globally　feed-forward　neural　network　which　consists　of　static　nonlinear　and　dynamic　linear　subsystems,　and　its　dynamic　properties　are　realized　using　neurons　with　internal　feedback.　This　proposed　RNN　can　be　stated　that　if　all　neurons　in　the　networks　are　stable　which　is　guaranteed.　Finally,　compared　with　the　normal　feed　forward　networks,　the　experiment　results　show　that　this　proposed　RNN　is　more　efficient　in　modeling　the　wastewater　treatment　system.