The moisture content of citrus is one of the key factors affecting the subsequent storage and processing of citrus. In order to detect the moisture content of citrus, the visible/near infrared transmission spectroscopy was used to detect the moisture content of satsuma orange. Differential processing, multivariate scattering correction, standard normal variate,SG convolution smoothing and MinMaxScaler are used and compared. At the same time, the Competitive adaptive reweighted sampling algorithm was used to extract the characteristic wavelengths, and then the partial least squares regression model(PLS), BP neural network model and least squares support vector machine (LSSVM) model based on citrus moisture content were established. The results show that the LSSVM model with 359 wavelengths obtained by CARS screening using the SNV preprocessed spectrum is the best predictor. The correlation coefficient and root mean square error of the correction set are 0.9375 and 0.0086, respectively. The square root errors are 0.8316 and 0.0120, respectively. The results show that the visible/near infrared spectroscopy technique is feasible to detect the water content of satsuma orange.