Every　year,　billions　of　films　appear　in　the　box　office　of　the　mainland　but　there　is　small　statistics　of　sample　data　for　them.　There　are　numerous　factors　responsible　for　it　e.g.　complex,　variable　box　office　elements　and　low　accuracy　of　box　office　demand　forecasting.　Whereas,　partial　least　squares　regression　model　has　the　capability　to　deal　with　small　sample　data　and　variable　multiple　correlations.　This　paper　has　conducted　an　empirical　analysis　by　using　13　indexes　affecting　the　movie　box　office　to　construct　movie　box-Office　forecast　model　as　well　as　analyze　the　principles　and　the　construction　steps　of　the　models.　The　model　has　utility　with　respects　to　process　and　model　accuracy.　The　empirical　results　show　that　the　absolute　relative　error　of　the　partial　least　squares　regression　model　is　26.6%,　the　goodness　of　fit　is　87.7%.　It　shows　that　the　partial　least　squares　model　has　great　skills　to　demonstrate　the　prediction　of　results　in　accurate　and　fashioned　way.