This　paper　proposes　a　data-driven　soft-sensing　method　for　predicting　effluent　ammonia　nitrogen　(NH4-N)　in　the　wastewater　treatment　process　(WWTP).　In　this　method,　a　rule　automatic　formation-based　adaptive　fuzzy　neural　network　(RAF-AFNN)　is　designed.　The　RAF　algorithm,　which　consists　of　rule　self-splitting　strategy　and　fuzzy　Gaussian　kernel　clustering,　is　used　to　automatically　partition　the　input　space　and　adaptively　extract　the　most　suitable　fuzzy　rules.　An　improved　adaptive　Levenberg-Marquardt　learning　algorithm　is　implemented　to　tune　the　parameters　of　the　RAF-AFNN　for　improving　prediction　accuracy.　An　analysis　of　the　convergence　is　also　provided　in　this　paper,　which　can　guarantee　the　successful　application　of　the　proposed　RAF-AFNN.　Finally,　experimental　hardware,　constructed　from　an　online　sensor　array　and　via　the　soft-sensing　method,　is　used　to　assess　the　effectiveness　of　the　RAF-AFNN　for　solving　the　problem　of　effluent　NH4-N　prediction　in　the　WWTP.　Experimental　results　indicate　that　the　proposed　RAF-AFNN-based　soft-sensing　method　can　predict　the　effluent　NH4-N　precisely.