The　octane　number　of　gasoline　was　considered　as　a　function　of　paraffin　lump,　naphthalene　lump,　aromatics　lump　and　olefins　lump　based　on　lumping　concept　for　complex　reaction　kinetics.　Back-propagation　(BP)　neural　network　and　multiple　linear　regression　were　adopted　to　establish　prediction　models　for　the　research　octane　number　(RON)　of　clean　gasoline　obtained　from　secondary　reactions,　respectively.　The　two　models　were　compared　and　verified　via　several　cases.　The　results　show　that　BP　neural　network　model　exhibits　better　performance　than　multiple　linear　regression　model　for　the　higher　prediction　accuracy　due　to　strong　nonlinear　mapping　ability　to　reflex　the　complex　relationship　between　RON　and　lump　components.　The　mean　absolute　relative　error　between　predicted　gasoline　RON　and　experimental　data　gets　0.39%,　and　0.92%　for　the　reported　results.