Soluble solids content (SSC) is the key indicator to evaluate the quality of watermelon pulp. In order to meet the needs of different groups of people and improve market competitiveness, an online detection model of watermelon SSC is established, which can realize the online grading of watermelon quality according to its SSC. In this paper, the 160 Jingmei2K watermelons are used as the research object, and the visible near-infrared full transmission spectrum data of the two postures of watermelons are collected using the online detection equipment independently developed by our laboratory. The partial least squares regression (PLSR) prediction model is established with the SSC of different parts of the watermelon to explore the best posture and part of online detection of watermelon SSC.Firstly, the SSC measurements of different parts of watermelon were defined as Pedicel Sugar, Central Sugar, Melon Navel Sugar and Average Sugar, and the two postures detected online were defined as T1 posture and T2 posture, respectively.Secondly, comparing the SSC of different parts of watermelon, the evaluation standard of watermelon SSC was discussed. Then, the spectral data with low transmission intensity and high frequency containing much noise and useless information were removed. Finally, the spectrum with a wavelength range (671~1 116 nm) was selected for analysis. The Savitzky-Golay smoothing (SGS) algorithm is combined with multiplicative scatter correction (MSC), unit vector normalization (UVN) and standard normal variate transformation (SNV) to preprocess the spectral data under two postures. Then the prediction model is established for the SSC of different parts of watermelon. By comparing the prediction results of different models, it is found that the combination of SGS and MSC has the best preprocessing effect for T1 posture spectral data, while The spectral data of T2 posturehas better performance using SGS combined with UVN preprocessing methods. The prediction effect of the T1 pose is better than that of the T2 posture spectral data. The prediction results of Pedicel Sugar and Average Sugar are better than that of Melon Navel Sugar, and Central Sugar is the worst. Finally, competitive adaptive reweighted sampling (CARS) was used to optimize the prediction models of Pedicel Sugar and Average Sugar. 81 and 106 wavelength points were selected to establish the prediction model of Pedicel Sugar and Average Sugar, respectively. The correlation coefficients of the prediction sets of the two models are 0.881 0 and 0.875 8, and the root mean square errors are 0.866 7% and 0.758 9%, respectively, simplifying the model and improving the prediction accuracy.The results showed that different postures and SSC prediction of different parts of a watermelon affected the results of online detection and quality evaluation. The model should be selected and optimized according to the actual needs of users. This paper, proposes an evaluation index for the online watermelon SCC detection, which provides a technical basis for further development of watermelon SSC online detection equipment.