Abstract: Fatty acids were stable metabolites and easily accumulated in paddies mould process which could better express mould extend of paddies.To achieve the non-destructive and rapid detection in fatty acid contents (FAC) for mould paddies, the detection of FAC for mould paddies was studied using the Visible/Near-infrared reflectance(Vis/NIR) spectral technology.The variety C liang-you 34156 late rice was used as paddy samples, which was obtained from Hunan Agricultural University.The mould paddy cultivating test and FAC determination experiments were carried out from October 15, 2014 to January 31, 2015 in Central South University of Forestry and Technology.Normal and complete paddies were selected and loaded into 200 sample boxes by numbers.Each sample box was loaded with 100g weights.Among them, 50 sample boxes were put into the No.A constant temperature humidity chamber to store according to requirements of cereal storage (temperatures:10 ℃, humidities:15%) and the remaining 150 sample boxes were placed in the No.B constant temperature humidity chamber to cultivate according to mould paddies breeding conditions (temperatures: 30 ℃, humidities: 90%).In view of the FAC variations affected by degree of mould paddies, the cultivated process of mould paddies was divided into three periods for better representative and generalization of samples.It was 10 days in each period, and 50 pieces of mould paddy samples in different degrees were measured during the preparation process.The Vis/NIR-infrared spectral detection testing for mould paddy samples were performed in corresponding periods in South China Agricultural University.A Vis/NIR-infrared spectral device for agro-products was used for scanning of reflectance spectra for paddies.Taking into consideration that the disadvantage of time consumption and low precision in building the model, the Vis/NIR calibration model of the fatty acid in mould paddies was proposed using sample set partition based on joint X-Y distances(SPXY)algorithm in sample set.The difference of predicting FAC in mould paddies from different calibration set was preliminarily analyzed using the combination of the SPXY algorithm and the partial least-squares regression (PLSR) algorithm.The successive projection algorithm(SPA) was applied to obtain the characteristic wavelength which indentified the variation of FAC in mould paddies.The predicted models of the FAC in mould paddies based on reflection values of characteristic wavelengths were built using the PLSR and multiple linear regression (MLR) methods, and then the prediction performance were compared between the model built by the selected calibration sample set and the model built by initial calibration sample set.The results indicated that FAC of paddies which were determined from different stages had a varying gradient distribution.The related FAC from the normal stage, early stage of mould, middle stage of mould and last stage of mould ranged from 18.55 to 24.40 mg, from 27.03 to 80.90 mg, from 84.44 to 127.26 mg, and from 101.09 to 124.88 mg, respectively.The range of FAC in 65 calibration sample sets by the SPXY was consistent with in 155 initial calibration sample sets.The standard deviation of FAC in 65 calibration sample sets was 32.39, which was close to the initial calibration sample sets.Nine characteristic wavelengths were selected from 256 full wavelengths by the SPA, which fulfilled the spectral data compression.The prediction set correlation coefficient (Rp) of the SPXY-SPA-PLSR model and the SPXY-SPA-MLR model were 0.922 1 and 0.915 9 and their prediction mean square root errors were 13.889 3 and 14.261 0, respectively.The model prediction precision built by the SPXY calibration set was close to its by the initial calibration, while the number of the SPXY calibration set was dropped to 41% and its computing time was reduced to 32% compared with the initial calibration, which may contribute to speed up the model establishment.