The　prediction　of　fixation　on　traffic　signs　was　conducted　based　on　image　characteristics.　The　fixation　prediction　model　which　is　based　on　a　model　of　visual　attention　was　set　up　and　improved　through　the　process　of　Gaussian　pyramid　creation,　visual　feature　extraction　and　conspicuity　map　generation,　as　well　as　the　saliency　map　generation.　The　model　was　validated　with　the　aid　of　an　eye　tracking　system.　Similarity　and　linear　distance　were　selected　as　two　indicators　to　evaluate　the　accuracy　of　model.　The　experimental　results　show　that　the　model　has　a　good　accuracy　on　the　prediction　of　traffic　sign　with　both　letters　and　symbols.　©　2011　ASCE.