The feasibility of Fourier transform-near-infrared (FT-NIR) spectroscopy for predicting the extractable content of phenolic compounds directly in intact grape seeds was evaluated. NIR calibration models were based on the correlation of spectral data with the phenolic composition determined by reference chemical methods for 40 grape samples. The vintage effect was studied evidencing that the predictive accuracy improved only for spectrophotometric indices when samples of two years were simultaneously considered. The statistical parameters of calibration showed that the models developed are sufficiently robust for quantitative purposes of total flavonoids, proanthocyanidins, low molecular weight flavanols, catechin, epicatechin, procyanidin B1, and galloylation percentage (standard prediction error (SEP%) < 15, predictive index (RPIQ) > 3.0), but could be also useful for screening of absorbance at 280 nm, total polymers, epicatechin gallate, and procyanidin B2 (SEP% < 15, RPIQ = 2.7-2.9). Although a calibration model is required for each geographical origin, the results suggest that FT-NIR spectroscopy is a promising analytical technique in this field.