In　order　to　solve　the　problem　of　controlling　dissolved　oxygen　(DO)　concentration　and　nitrate　concentration　of　waster-water　treatment　process　(WWTP),　a　multi　critic　indices　dual　heuristic　dynamic　programming　(MDHP)　policy　is　proposed.　The　approximating　precision　can　be　improved　through　lowering　the　complexity　between　the　relationship　of　the　critic　network＇s　outputs　and　inputs　in　this　scheme.　Echo　state　networks　(ESNs)　are　adopted　to　approximate　the　critic　indices　and　the　optimal　control　policy.　Online　learning　method　of　the　controller　is　investigated.　Experimental　results　indicate　that　the　MDHP　scheme　has　some　advantages　over　single　critic　index　DHP　(SDHP)　and　PID　in　control　performance.　Copyright　©　2013　Acta　Automatica　Sinica.　All　rights　reserved.