Fitting　curve　is　a　critical　problem　in　many　testing　equipment　and　detection　system.　But　there　was　larger　relative　error　in　fitting　curve　when　the　independent　variable　was　relatively　small.　In　this　paper,　Fitting　curve　is　formulated　to　a　constrained　linear　programming.　a　neural　dynamics　optimization　algorithm　is　obtained　by　considering　the　problem　in　its　dual　space,　and　then　the　dynamic　neural　network　is　designed　to　solve　the　optimization　problem　recurrently.　The　experimental　results　show　that　the　polynomial　coefficients　solved　by　the　method　is　stable,　compared　with　the　least　square　method,　the　relative　error　is　obviously　reduced;　The　method　is　simple　and　requires　less　samples.　It　provides　a　new　simple　and　accurate　method　of　curve　fitting　for　the　quantitative　detection.　©　Springer　International　Publishing　AG　2017.