heat seed germination rateis one of the key index for evaluating theseed quality which is the basic factor determining the yield and quality of wheat.Germination test to determine germination rate in national standard is time-consuming(4-8days) and fussy operation thus can't adapt tothedevelopment of the modernagriculture needs.In recent years, near infrared spectroscopy (NIR) technology has provided new thought and method for the determining of seed germination rate with its rapid, non-destructive, accurate and non-pollution etc.advantages.Determining seed germination rate with NIR was littleresearched currently and mainly relied on establishing grayquantitative analysis model associated NIR spectral withgermination rate,however no further explores reported on the modeling mechanism.This studyaimed at selecting the characteristic spectral regions of wheat seed germination rate via two different representative variable selection methods, including backward interval partial least squares (BiPLS) and synergy interval partial least squares(SiPLS),and analyzing the physical and chemical indicesaffecting wheat seed germination rate thoughthe interpretation of characteristic spectral regions, which can lay the theoreticalfoundation for establishing good NIR quantitative analysis models of wheat seed germination rate.SiPLS split full NIR spectral region (4000-12000cm) into 5-35 intervals combined in up to 5 subintervals to develop PLS model and selected the best results.The optimal model was obtained when divided 11 spectral regions and 2 intervals(5450-6172cm-1 and 6901-7622cm-1)were selected with 7 PLS factors, root mean square error of cross-validation(RMSECV)at 0.1608, root mean square error of prediction(RMSEP) at 0.1063, calibration and prediction correlation coefficient(R) at 0.869 and 0.9588, and ratio of performance to standard deviate (RPD) at 3.49.While BiPLS was implemented with the full spectrum divided into 5-40 intervals, the interval combination with lowest RMSECV was regarded as final model interval.The optimal model was obtained when divided 9 spectral regions and 3 intervals(5774-6658cm-1,6662-7545cm-1,8436-9323cm-1)were selected with 7 PLS factors,RMSECV at 0.1520, RMSEP at 0.1172, calibration and prediction correlation coefficient(R) at 0.882 and 0.9534, and RPD at 3.17.Comparison results showed that the models built by two variable selection methods had higher predictive ability and more simplify than full-spectral model Further this studycomparedand analyzed the spectral region selected by two variable selection methods found that they have overlap regions at 5774-6172cm-1 and 6901-7545cm-1 which including doubled-frequency and combined-frequency absorption of C-H in CH3 or CH2, the first-order doubled-frequency absorption for stretching vibration of O-H in liquid water (6944 cm-1)and the non-bonded hydroxyl(-OH) absorption(7065cm-1).C-H exist in organic matter such sugars, fiber,protein which are the major ingredient in seed and provide nutrition for seed germination.The absorption of O-H in spectral is related to free water in seedwhose contentis one of the key factors control whether seed germination.Besides, spectral regions selected by BiPLS also included the first-order doubled-frequency absorption of free N-H (6710-6500cm-1) which may exist in free-amino acid, while free-amino acid is an important symbol to measure quality changes of seed for the reason that protein will hydrolyze into free-amino acid when in the ageing process or if the storage environment is unsuitablewhich can causeseed moisture absorption and germination ability reduced.All these showed that Wheat seed germination ratehad been associated withNIRessentially originated from NIR characteristic absorption of hydrogen-containedgroups in seed component.Correlation and influence of each component in seed to germination rate and NIR prediction modelneed to be further investigated.