The　method　of　evaluating　pavement　roughness　with　vibration　acceleration　data　collected　by　smart　phones　is　influenced　by　many　factors,　of　which　the　length　of　pavement　evaluation　unit　is　an　important　one.　In　order　to　explore　the　influence　of　evaluation　unit　length　on　pavement　roughness　evaluation　result　with　vibration　data,　this　paper　carried　out　a　driving　experiment,　used　the　self-developed　smart-phone　APP　acquisition　running-vehicle　vibration　of　data,　used　the　multi-function　checkout　automobile　detection　of　pavement　smoothness,　and　established　the　relationship　models　of　pavement　roughness　-　vibration　index　and　speed　index　when　the　evaluation　unit　length　are　100　m,　500　m　and　1000　m　respectively.　The　results　showed　that　when　the　unit　length　is　1000　m,　the　accuracy　of　the　evaluation　model　is　the　highest,　and　the　determination　coefficient　reaches　0.876.　The　verification　experiment　also　proved　that　the　evaluation　model　has　good　accuracy　and　stability.　And　also　it　was　showed　that　the　accuracy　of　the　pavement　roughness　evaluation　model　increases　with　the　increase　of　the　length　of　the　evaluation　unit,　which　may　be　related　to　the　matching　error　between　the　vibration　data　and　the　pavement　roughness.　How　to　improve　the　accuracy　of　the　evaluation　results　about　the　short　unit　is　the　key　content　to　be　studied　in　the　future.　©　Published　under　licence　by　IOP　Publishing　Ltd.