A　method　for　urban　major　road　information　extraction　from　IKONOS　imagery　is　proposed.　Firstly,　a　gradient　algorithm　based　on　vector-valued　images　is　used　to　calculate　gradients　of　the　IKONOS　image　of　Fuzhou　city　with　four　multi-spectral　bands.　Secondly,　spectral　signatures　of　road,　bare　soil,　building,　vegetation　and　water　from　the　IKONOS　image　are　analyzed,　and　it　follows　that　the　road　can　be　effectively　distinguished　from　bare　soil,　vegetation　and　water　in　the　differencing　image　between　green　and　near-infrared　bands.　In　addition,　the　road　variance　induced　by　different　constructional　materials　is　reduced　in　this　differencing　image　than　in　the　original　multi-spectral　bands.　In　order　to　further　distinguish　between　road　and　building,　a　rotation-invariant　Gabor　wavelet　method　is　used　to　obtain　texture　features.　Accordingly,　the　gradient　and　spectral-textural　features　of　the　road　are　integrated　to　develop　a　velocity　function.　Finally,　the　fast　marching　algorithm　is　used　to　extract　major　roads　and　a　mathematical　morphology　method　is　used　to　optimize　the　above　extraction　results.　This　proposed　method　is　used　to　extract　major　roads　from　two　test　areas　of　Fuzhou　city　with　different　cloud　conditions.　The　experiment　results　show　that　this　method　is　effective　to　extract　major　roads　from　IKONOS　imagery　even　with　thin　cloud　coverage.