Nondestructive determination of grape ripeness is essential for vineyard harvest schedule optimization. This study aims to investigate the identification of less ripen, ripen and over-ripen grapes by visible and near infrared (Vis-NIR) spectrophotometer. A local cultivar of grape ‘Beyond’ was examined during harvest season from July to August, 2011. The samples were separated into three ripen degrees, e.g. less-ripen, ripen, and over-ripen, according to the sugar content in grapes. The samples were divided into a calibration set (70%) and an independent prediction set (30%). The calibration set was subjected to a partial least-squares (PLS) regression and principal components analysis (PCA) with leave-one-out cross validation. The first 10 factors, e.g. latent variables (LVs) for PLS and principal components (PCs) for PCA, were chosen as input variables to three classification models, e.g. linear discrimination analysis (LDA), back-propagation artificial neural network (BPANN) and support vector machine (SVM). These models were validated by the independent prediction set. Validation result shows that PCA-LDA and PLS-LDA models achieve higher classification accuracy than others. The LDA combined with 6 PCs performs best with 100% classification accuracy. It is concluded that Vis-NIR spectroscopy is promising for the instant identification of different ripeness of grapes. The proposed technique is useful for discriminating ripen and over-ripen grapes during harvest time.