Objective: To reduce the influence of transverse diameter of Mandarin orange on the prediction of sugar content. Methods: 132 citrus were divided into three groups according to transverse diameter: small （65～70 mm）, medium （70～75 mm） and large （75～80 mm）. After collecting the transverse diameter spectrum of all citrus, the spectrum information and transverse diameter information of citrus were synthesized by spectral transformation algorithm, and the spectrum of different sizes of citrus was converted to the same transverse diameter datum. The pre-correction and post-correction spectra were respectively preprocessed, divided by target symbiotic distance algorithm （SPXY）, screened by competitive adaptive weight sampling （CARS） and selected by partial least squares regression （PLS） to establish the sugar degree model before and after correction. Results: The prediction set determination coefficient （R²_P） was increased from 0.790 to 0.821, and the prediction set root mean square error （RMSEP） was decreased from 0.489 to 0.443. The middle fruit spectrum was modified, R²_P increased from 0.801 to 0.845, RMSEP decreased from 0.460 to 0.422. R²_P increased from 0.820 to 0.863 and RMSEP decreased from 0.431 to 0.393. Conclusion: The spectral correction algorithm can reduce the spectral difference caused by the transverse diameter and improve the prediction accuracy of the model.