Considering that the current methods of germination rate detection are complex, time-consuming and affected by seed dormancy, a method for rapid detection of brown rice germination rate based on characteristic spectrum and general regression neural network (GRNN) is proposed. Under the condition of temperature 45 ℃ and relative humidity 90%, rice seeds are aged artificially for 0, 24, 48, 72, 96, 120, 144, 168 h. Spectral data of 160 samples are collected by a near-infrared spectrometer after artificial shelled processing and divided into a calibration set (120 samples) and a prediction set (40 samples). Characteristic wavelengths are extracted after standard normalized variate (SNV) and first derivative (FD) preprocessing. The impact of different modeling methods and characteristic wavelengths on the model is analyzed. The experimental results show that the optimal model is constructed by GRNN with the spectral data of 688, 1146, 1346, 1366, 1396, 1686 nm. The correlation coefficients of the calibration set (R_C) and the prediction set (R_P) are 0.9743 and 0.9505, and the standard errors of the calibration set (SEC) and the prediction set (SEP) are 1.9161 and 2.3423. The research results show that it is feasible to measure the germination rate of brown rice seeds by using near infrared spectroscopy. The model has better predictive ability in germination rate, and reveals the difference between rice seeds with different germination rate from the perspective of physiological characteristics. This method provides a theoretical basis for the development of portable spectrometer for rice seed germination rate detection.