Abstract: In order to solve the insufficient timeliness in the detection of phenolic contents in fresh rice bran, a rapid nondestructive detection method by using Fourier transform near infrared spectroscopy (FT-NIR) was constructed in this work. Multiple batches of fresh rice bran were utilized as experimental samples. Contents of free, bound and total phenolic compounds in fresh rice bran were quantitatively detected, and the prediction models of partial least squares regression (PLSR), support vector machine (SVM) and back propagation neural network (BPNN) were established based on full and characteristic wavelengths of FT-NIR. The results showed that the models based on PLSR algorithm obtained the best predicted performance based on full-wavelength datasets, the R_\rmp^2 reached 0.944, 0.943 and 0.937, and RPD reached 3.031, 2.779 and 2.863 for free, bound and total phenolic compounds, respectively. Two variable selection methods, named competitive adaptive reweighted sampling (CARS) and continuous projection (SPA) algorithms, were used in this work, and several key wavelengths from 4 to 8 numbers were selected. Best predicted models using key-wavelength datasets were constructed by CARS-PLSR, with R_\rmp^2 of 0.953, 0.932 and 0.944 for bound, free and total contents, respectively. The RPD could obtain with 3.301, 2.759 and 3.031, respectively. Meanwhile, the running time of modeling was shortened by 2 times and only needed 2.0 s, which could meet the timeliness requirement of detection of phenolic compounds in rice bran. The results of this study confirmed that FT-NIR technique could be used for rapid and quantitative determination of phenolic components in rice bran.