Based　on　principal　component　analysis　neural　network,　within　a　probabilistic　framework,　we　introduce　a　new　Monte　Carlo　tracking　technique　for　autonomous　navigation　of　land　vehicle　on　unpaved　road.　The　use　of　straight-road　model　and　particle　filter　allows　us　to　handle　blurry　road　boundaries,　and　the　use　of　color　space　transform,　local　spatial　features　and　principal　component　analysis　facilitates　adaptation　to　the　road　conditions.　Experimental　results　verify　the　algorithm　in　different　conditions.　©　Springer-Verlag　2004.