The ability to estimate the sensory quality of intact tomatoes rapidly and non-destructively using visible and near-infrared spectroscopy (Vis-NIRS) is important for the tomato industry. In this study, a combination of partial least squares regression (PLSR) analysis and the stepwise selectivity ratio (SWSR) method was used to study the ability of Vis-NIRS to predict 19 sensory attributes in intact tomatoes. The PLSR models constructed based on the informative wavelengths selected by the SWSR method predicted 8 sensory attributes well, particularly the sweetness attribute (correlation coefficient of validation of 0.92). Moreover, based on the tomato metabolites determined by GC-MS analysis, high intercorrelations between sensory attributes, metabolites, and the selected informative wavelengths were found through principal component analysis, as well as the high correlation coefficients between them. The results confirm the feasibility and reliability of Vis-NIRS and the informative wavelengths selected by SWSR to predict the sensory quality of whole tomatoes.