The potential of predicting translocatable matter of rice with near infrared reflectance spectroscopy (NIRS) was studied. Using 7 varieties of rice planted in Danzhou of Hainan province as materials, the method of neutral detergent fiber added amylase with NIRS was examined to establish calibration model of predicting translocatable matter of stem and panicle of rice. The results indicated that partial least square(PLS1) is the best regression statistic method for calibration model; The differences of results of the spectral data pretreatment methods for calibration model were insignificant; Because of the high prediction accuracy, the final calibration model was chosen using “no spectral data pretreatment”+“PLS1”; Determination coefficient of external validation and root mean square errors of prediction of the calibration model of stem and panicle was 0.991 2, 0.008 1, 0.961 1 and 0.022 6, respectively.