Macroscopic　traffic　flow　models　play　a　fundamental　role　in　traffic　state　estimation　and　prediction,　as　well　as　traffic　control　engineering.　The　cell　transmission　model　(CTM)　has　been　thought　as　an　approbatory　one　capable　of　practical　applications　in　predicting　traffic　state.　To　improve　the　accuracy　of　the　CTM,　this　paper　proposes　a　modified　cell　transmission　model　(MCTM)　by　using　feed-forward　neural　network　to　describe　nonlinear　transfer　flows　between　cells.　We　adopt　the　feed-forward　neural　network　with　trained　parameters　by　historical　data　from　induction　loop　detectors　to　improve　the　accuracy　of　transmitted　flow　expressions.　Then　we　compare　performance　of　this　MCTM　combined　with　neural　network　method　with　that　of　the　primitive　MCTM　by　building　a　VISSIM　model　of　a　segment　of　Jitong　Freeway.　The　simulation　results　show　that　the　proposed　model　has　a　more　optimal　performance　on　predicting　traffic　density　than　the　primitive　model.　©　2018　American　Society　of　Civil　Engineers.