Objective To lower the complexity, improve the accuracy and reduce the damage of the detection process of citrus sugar content.Methods An online detection device for citrus sugar content is designed based on visible near-infrared reflectance spectroscopy. With Jinniu Citrus as the research object, the modeling set and verification set are divided by sample set partitioning based on the joint x-y distance (SPXY) classification method. The partial least square regression (PLS) modeling and detection effects after pretreatment are respectively compared and analyzed by methods including multiple scattering correction (MSC), standard normal variation (SNV), and SG-smoothing (SG) to determine the optimal pretreatment method. At the same time, a comparative study is conducted on the extraction of feature bands from pretreatment spectral data using the successive projections algorithm (SPA), the competitive adaptive reweighted sampling (CARS), and the random frog (RF) algorithm. Suitable feature wavelength points are screened out and the PLS prediction models are established.Results The PLS model established by screening out the 95 feature wavelength points using SG+MSC+CARS has the best prediction performance. Its correlation coefficient of calibration (R_c) and correlation coefficient of prediction (R_p) are 0.913 and 0.881, respectively, root mean square error of the calibration set (RMSEC) and root mean square error of the prediction set (RMSEP) are 0.274 and 0.207, respectively, and residual predictive deviation (RPD) is 2.114.Conclusion This method effectively lowers the complexity of the citrus sugar content detection process, improves the detection accuracy, and reduces the detection damage.