This　paper　presents　a　structure-adaptive　median　filter　for　reducing　random　seismic　noise　which　preserves　seismic　edges　and　details　such　as　faults　and　fractures.　By　considering　seismic　horizons　as　line-like　structures,　this　filtering　framework　relies　on　the　computation　of　the　Gradient　Structure　Tensors,　which　provides　dips　of　seismic　events　and　the　regularity　of　local　seismic　structures,　based　on　which　two　confidence　measures　are　defined　for　different　seismic　structures.　The　structure-adaptive　median　filter　adjusts　the　shape　of　the　filter　kernel　according　to　the　two　confidence　measures　and　aligns　the　filtering　kernel　along　local　dips　to　make　the　filtering　kernel　optimally　matched　with　different　local　geological　features.　The　proposed　filter　has　been　applied　to　both　the　synthetic　and　real　data,　and　compared　with　two　widely　used　median　filtering　schemes.　The　processing　and　comparison　results　show　that　the　proposed　structure-adaptive　median　filter　is　more　suitable　to　balance　suppressing　random　seismic　noise　and　preserving　signals　which　preserves　seismic　edges　and　details　while　enhancing　the　coherence　of　seismic　horizons,　and　improves　the　quality　of　seismic　images　significantly.