Short-term　traffic　flow　forecasting　plays　a　significant　role　in　the　Intelligent　Transportation　Systems　(ITS).　Previous　research　shows　that　the　successful　predictability　decreases　sharply　in　multiple-steps-ahead　prediction　due　to　the　fact　that　freeway　traffic　state　changes　abruptly　under　heterogeneous　traffic　conditions.　This　paper　proposes　a　pattern-based　forecasting　framework　that　is　consistent　with　the　evolution　of　traffic　flow　and　has　the　ability　to　exploit　past　traffic　pattern　information　in　order　to　enhance　predictability.　For　the　purpose　of　prediction,　several　historical　traffic　patterns　are　developed　for　each　traffic　state　according　to　the　classified　historical　traffic　data.　The　future　value　of　traffic　flow　can　be　quickly　predicted　with　the　real-time　traffic　data　by　means　of　pattern　matching　techniques　to　identify　the　current　traffic　pattern.　Finally,　a　comparative　study　on　a　section　of　the　Third-Loop　Freeway,　LIULIQIAO,　Beijing,　indicates　that　pattern-based　forecasting　that　considers　traffic　state　is　more　accurate　when　compared　to　classical　ARIMA　models,　which　only　operate　under　the　time　series　consideration.　©　2010　ASCE.