The　travel　speed　prediction　method　proposed　in　this　paper　takes　advantages　of　fuzzy　inference　systems　that　are　insensitive　to　random　noise,　robust　to　uncertainties,　and　transparent　to　interpretation.　The　collected　traffic　data　is　pre-processed　by　removing　the　outliers　based　on　the　density　distribution　of　the　raw　data.　Subsequently,　a　clustering　is　performed　on　the　pre-processed　data　to　construct　a　Gaussian　mixture　model　and　this　mixture　model　is　then　used　to　identify　the　parameters　for　the　fuzzy　inference　system.　A　comparative　evaluation　of　the　proposed　prediction　method　was　conducted　using　the　real　traffic　data.　The　results　indicate　that　the　proposed　prediction　method　gives　an　improved　performance.　©　2010　ASCE.