Process　modeling　plays　a　very　important　role　in　process　system　engineering.　However,　it　is　more　and　more　difficult　to　develop　an　accurate　model　for　a　complex　chemical　process　due　to　the　highly　nonlinear　and　complicated　process　data.　For　the　purpose　of　handling　this　problem,　an　improved　functional　link　learning　machine　using　singular　value　decomposition　(SVD-FLLM)　is　presented　to　develop　accurate　models　for　complex　chemical　processes.　In　the　proposed　SVD-FLLM　model,　singular　value　decomposition　is　adopted　to　reduce　the　expanded　variable　dimension.　Then,　the　least　square　algorithm　is　used　to　build　a　regression　model　between　the　reduced　outputs　and　the　expected　vectors.　For　the　sake　of　validating　the　testing　effectiveness　of　the　proposed　SVD-FLLM　method,　a　UCI　dataset　named　Airfoil　Self-Noise　is　first　selected.　Then　the　proposed　SVD-FLLM　is　used　to　develop　a　model　for　an　actual　complex　chemical　system.　Simulation　results　indicate　that,　compared　the　traditional　functional　link　neural　network　(FLNN),　the　improved　SVD-FLLM　can　achieve　higher　accuracy　and　faster　convergence.