This study compared the stability and accuracy of the portable near-infrared spectrograph (901~1 650 nm) and visible spectrograph (400~900 nm) in nondestructive detection of moisture content in rice. 100 different varieties of rice were selected and their spectral information was collected, including japonica rice, indica rice and glutinous rice. The number of varieties were 52, 34, 14. Firstly, the direct drying method in GB 5009.3—2016 was used to determine the water content of each rice sample. Then, the outliers in rice samples were eliminated by Monte-Carlo partial least squares method, 8 and 4 outliers were eliminated from the dataset based on near-infrared and visible spectra. Moreover, the sample set partitioning based on joint x-y distance (SPXY) was used to divide the samples according to 3∶1. The near-infrared and visible data sets obtained 69, 72 calibration sets and 23, 24 prediction sets, respectively. In addition, nine algorithms including orthogonal signal correction (OSC), multivariate scattering correction (MSC), de-trend, standard normal variate (SNV), baseline, Savitzky-Golay convolution derivative (S-G derivative), normalization, moving average smoothing, and Savitzky-Golay convolution smoothing (S-G smoothing) were used to preprocess the original spectral data, OSC and SNV based on near-infrared and OSC and moving average based visible spectra had good effects, and subsequent model processing is carried out. Finally, feature wavelengths were selected to reduce spectral information redundancy and improve the model detection effect. The best wavelength selection methods based on near-infrared and visible spectra were successive projections algorithm (SPA) and competitive adaptive reweighting sampling (CARS), respectively, with 15 and 39 feature wavelengths reserved. And then, partial least squares regression (PLSR) and principal component regression (PCR) models were established. The results showed that the best combination of the models for near-infrared and visible spectra were SPA-PLSR and OSC-CARS-PCR, respectively. The correlation coefficient (R2P), root mean square error (RMSEP) and normalized root mean square error(NRMSEP) of the prediction set were 0.810 3, 0.802 1, 0.412, 0.388 and 3.62%, 3.34%, respectively. The SPA-PLSR model based on the near-infrared spectrum had a better prediction effect, and better robustness than other models. The prediction effect of the near-infrared spectrum was better than that of the visible spectrum. This study verified the feasibility of portable near-infrared spectrograph and visible spectrograph for rapid and nondestructive detection of moisture content in rice, provided technical support for determining moisture content in rice harvesting, storage and other processes, and provided a reference for the development of subsequent portable spectrographs.