The sensory preference of consumer was influenced by the internal qualities of fruit, which directly determined the tendency of consumer’s purchase. The visual-near infrared spectroscopy was employed to inspect the internal qualities of “kyoho” table grape and the grade of the sensory preference, aiming to provide the fruit with superior qualities for consumers. Spectra in diffuse reflection and transmission acquisition mode were comparatively conducted to measure the internal qualities of table grape. Results show that the transmission spectra had a better capacity to characterize the internal information of grape berries, and the partial least square (PLS) model was optimized to measure sugar content (SC) and total acidity (TA) with root mean square error of prediction (RMSEP) of 0.598%brix, 0.048 g·L-1 respectively. Three nonlinear classification methods were comparatively investigated between the sensory preference of customers and the principal component of transmittance spectra of table grape, and the extreme learning machine combined with principal component analysis (PCA-ELM) model obtained the best performance with the classification accuracy of 78.7%. It was concluded that spectra characterized the internal information of fruit could be used to preliminarily grade the sensory preference of consumers, however, the relationship between them requires further study.