When using near-infrared diffuse transmission technology to detect SSC of apple, the information of soluble solids in the collected apple spectrum is different due to the different positions of halogen lamp on apple, which will lead to the different performance of the model. Finding the best illumination position of apple is conducive to obtaining the best evaluation model of soluble solids. Using the multi-mode adjustable optical structure, the near-infrared diffuse transmission spectra of two batches of apples purchased from the same fruit wholesaler with similar size but different irradiation positions were collected under the same experimental environment and conditions. The best irradiation position in the process of establishing the apple soluble solid model was studied, and the evaluation model of the best position of soluble solid was obtained. The best modeling position is obtained by spectrum collection, true sugar degree value collection and chemometrics method. When the irradiation position is the upper part and the spectrum is not pretreated, the PLS(Partial Least Square) model performance is RMSEC 0.288 2, RMSEP 0.343 6, R_c 0.960 6 and R_p 0.934 9. The performance of the PLS model with oblique upper irradiation position and no spectral pretreatment is RMSEC 0.340 7, RMSEP 0.513 3, R_c 0.931 1 and R_p 0.863 6. The performance of the PCR (Principle Component Regression) model with upper irradiation position and no spectral pretreatment was RMSEC 0.573 6, RMSEP 0.601 4, R_c 0.842 4 and R_p 0.800 7. The performance of the PCR model with oblique upper irradiation position and no spectral pretreatment was RMSEC 0.709 2, RMSEP 0.797 4, R_c 0.701 4, R_p 0.670 7. The best irradiation position is the upper part of the apple; Further, a variety of pretreatment methods are used to compare the PLS model with the upper irradiation position. The optimal model is the MSC-PLS model. Its model performance is RMSEC 0.226 44, RMSEP 0.301 5, R_c 0.966 9 and R_p 0.949 9. Finally, after the same experimental operation is carried out on the same 46 apples, the spectra and true values are obtained and substituted into the established MSC-PLS (Multiplicative Scatter Correction-Partial Least Square) model for external verification. The results show that the correlation coefficient of external verification is 0.930 58, and the root mean square error of verification is 0.843 59, which verifies the stability and reliability of the established MSC-PLS model. It further shows that the near-infrared diffuse projection model has good prediction ability when the spectral acquisition position is the upper part of the apple. This paper provides technical support for predicting the detection of soluble solids in the apple.