As　an　important　assisted　feature　for　spectral　one,　texture　plays　an　important　role　in　image　analysis　and　automatic　recognition,　especially　in　high　spatial　resolution　remotely　sensed　images.　Meanwhile,　wavelet　is　an　effective　method　of　extracting　multi-scale　features　of　ground　objects　in　images.　Then　in　this　research,　the　extraction　approach　of　wavelet-domain　fractal　texture　(WDFT)　was　proposed,　and　it　was　implemented　to　improve　the　image　classification　of　QuickBird　of　Fuzhou　City.　WDFTs　of　QuickBird　image　were　computed　on　different　window　sizes　and　decomposition　layers,　and　three　texture　images　were　selected　from　the　viewpoint　of　image　classification　and　thematic　extraction　of　buildings,　the　different　box-counting　(DBC)　cap　features　of　CA1　(the　coarse　image　of　the　first　decomposed　layer　of　QuickBird　image)　on　64　*　64　and　16　*　16　windows　and　the　multi-fractal　feature　of　CA1　on　16　*　16　window.　The　experiment　results　implied　that:　because　of　the　addition　of　WDFT　information,　the　aquafarm,　major　roads　and　bare　land　confused　with　buildings　were　distinguished　well;　the　supervised　classification　based　on　spectral　feature　was　modified,　and　its　total　classification　accuracy　and　Kappa　coefficient　became　better　(from　76.17%　to　81.25%,　and　from　0.7006　to　0.7587,　respectively),　and　also　made　the　extraction　accuracy　(user　one　and　mapping　one)　of　buildings　better　(from　80.70%　to　82.54%,　and　from　65.71%　to　74.29%,　respectively).　It　proves　that　the　WDFT　was　effective.　Addressed　on　the　disadvantages　in　the　research,　the　WDFT　extraction　on　rectangular　window　and　adaptive　sizes　will　be　studied　and　more　decomposition　layers　information　will　be　integrated　in　the　next　work.