Abstract: In order to build a universal model of near infrared spectroscopy for determining oil content in three woody oil plant seeds in Hunan, 98 Vernicia fordii seed samples, 96 Camellia oleifera seed samples and 96 Juglans regia seed samples were collected. Near infrared spectra (NIR) of their crushed seed kernel were recorded. Oil content was determined. Partials quare least (PLS) and radical basis function neural networks (RBFNN) were used to develop the universal NIR models for determining oil content for each of 4 sample sets (i.e. V. fordii+C. oleifera+J. regia, V. fordii+C. oleifera, V. fordii+J. regia, and C. oleifera+J. regia), respectively. For PLS models, the correlation coefficient (Rp) were 0.972, 0.910, 0.980 and 0.981, root mean square error (RMSEP) were 2.44, 3.28, 2.04 and 2.49 and relative standard deviation (RSD) were 4.27%, 6.45%, 3.33% and 4.20% for validation sets of the 4 sample sets, respectively. For RBFNN models, Rp were 0.965, 0.894, 0.973 and 0.979, RMSEP were 3.04, 2.44, 2.32 and 2.27, RSD were 5.33%, 4.80%, 3.79% and 3.83% for them, respectively. The results showed that the universal models for determining oil content in three woody oil plant seeds could be built by using NIR technology.