Accurate　and　reliable　prediction　of　groundwater　level　is　essential　for　water　resource　development　and　management.　This　study　was　carried　out　to　test　the　validity　of　three　nonlinear　time-series　intelligence　models,　artificial　neural　networks　(ANN),　support　vector　machines　(SVM)　and　adaptive　neuro　fuzzy　inference　system　(ANFIS)　in　the　prediction　of　the　groundwater　level　when　taking　the　interaction　between　surface　water　and　groundwater　into　consideration.　These　three　models　were　developed　and　applied　for　two　wells　near　Lake　Okeechobee　in　Florida,　United　States.　10　years　data-sets　including　hydrological　parameters　such　as　precipitation　(P),　temperature　(T),　past　groundwater　level　(G)　and　lake　level　(L)　were　used　as　input　data　to　forecast　groundwater　level.　Five　quantitative　standard　statistical　performance　evaluation　measures,　correlation　coefficient　(R),　normalized　mean　square　error　(NMSE),　root　mean　squared　error　(RMSE),　Nash-Sutcliffe　efficiency　coefficient　(NS)　and　Akaike　information　criteria　(AIC),　were　employed　to　evaluate　the　performances　of　these　models.　The　conclusions　achieved　from　this　research　would　be　beneficial　to　the　water　resources　management,　it　proved　the　necessity　and　effect　of　considering　the　surface　water-groundwater　interaction　in　the　prediction　of　groundwater　level.　These　three　models　were　proved　applicable　to　the　prediction　of　groundwater　level　one,　two　and　three　months　ahead　for　the　area　that　is　close　to　the　surface　water,　for　example,　the　lake　area.　The　models　using　P,　T,　G　and　L　achieved　better　prediction　result　than　that　using　P,　T　and　G　only.　At　the　same　time,　results　from　ANFIS　and　SVM　models　were　more　accurate　than　that　from　ANN　model.