Aiming　at　features　of　traffic　structure　in　commercial　sites,　the　paper　discusses　the　factors　which　affect　the　public　traffic　modal　split.　The　BP　neural　network　can　approach　continuous　function　precisely.　So　under　the　strong　non-linear　mapping　and　generalization　characteristics　of　BP　neural　network,　a　prediction　model　of　commercial　sites　have　been　established　based　on　the　available　commercial　sites　survey　data　of　Beijing.　Meanwhile,　this　paper　has　developed　the　corresponding　program　taking　advantage　of　the　neural　network　toolbox　in　MATLAB.　When　the　factors　change,　it　can　quantitatively　analyze　the　influence　to　the　proportion　of　transit　trips.