The　traffic　noise　assessment　is　an　important　part　of　an　Environmental　Assessment　(EA)　for　highway　infrastructure　projects;　however,　its　accuracy　highly　depends　on　the　precision　of　traffic　noise　prediction　models.　In　this　study,　vehicle　noise　emission　values　are　expressed　in　terms　of　a　power　unit　component　and　a　rolling　noise　component.　First,　laboratory　experiments　are　conducted　to　collect　the　vehicle　speed　and　its　corresponding　power　unit　noise　emission　data.　Then　the　linear　model,　logarithmic　model,　and　quadratic　curve　model　are　developed　to　find　the　regression　fitting.　The　results　show　that　the　quadratic　curve　model　has　the　best　fitting　degree,　so　it　is　used　in　the　vehicle　noise　prediction　model.　Second,　the　rolling　noise　model　proposed　by　Hamet　and　Jean　-　Francois　is　introduced　to　develop　the　rolling　noise　emission　using　the　method　of　energy　superposition.　A　case　study　shows　that　the　proposed　revised　model　can　significantly　improve　the　accuracy　of　low　phase　noise　prediction.　In　general,　the　proposed　model　obtains　much　better　prediction　results　than　that　obtained　from　the　highway　construction　project　environmental　impact　assessment　guideline,　with　the　average　precision　increased　by　more　than　20%.　©　2017　IEEE.