In order to reduce the low accuracy of the prediction model of soluble solid content caused by the size difference of gongpear, a general size model was proposed. Near-infrared diffuse reflection spectroscopy and partial least squares regression algorithm modeling method of the near-infrared spectrum were adopted, and the theoretical analysis and experimental verification were respectively carried out. Gongpear soluble solids content of data were obtained by a local size prediction model of the small fruit, middle fruit and big fruit size grades and a universal size model with different size grade. The results show that, the local size model is good at predicting the soluble solid content of gongpears with its own grade, but poor at predicting other grades. The correlation coefficient were 0.892, 0.937, 0.889 and root mean square error of small fruit, medium fruit and large fruit predicted by the general model were 0.524, 0.417, 0.551. The general size model could reduce the adverse effects of size difference and was suitable for the determination of soluble solid content of gongpear with different size grades.