Polarimetric　SAR　remote　sensing　provides　an　outstanding　capability　of　oil　spill　detection　and　classification　for　its　advantages　in　distinguishing　mineral　oil　and　biogenic　look-alikes.　In　this　paper,　deep　learning　algorithms　including　Stacked　Auto-Encoder　(SAE)　and　Deep　Believe　Network　(DBN)　are　applied　to　optimize　the　polarimetric　feature　sets　and　reduce　the　feature　dimension　through　the　processes　of　layer-wise　unsupervised　pre-training.　An　experiment　was　conducted　on　RADARSAT-2　quad-polarimetric　SAR　image　acquired　during　Norwegian　oil-on-water　exercise,　in　which　verified　mineral,　emulsions,　and　biogenic　slicks　were　provided.　The　results　show　that　oil　spill　classification　achieved　by　deep　networks　outperformed　support　vector　machine　(SVM)　and　traditional　artificial　neural　network　(ANN)　with　similar　parameter　settings,　especially　when　the　number　of　training　data　samples　is　limited.