A 2(3)-full factorial design and response surface methodology were deployed to assess some basic factors (time, % ethanol and pH) affecting profoundly the extractability of polyphenolic phytochemicals from grape (Vitis vinifera) stems. In an effort to obtain a thorough insight into the applicability of the models established, stem extracts from three different varieties were tested, by determining several indices of the polyphenolic composition, such as total polyphenol (TP), total flavanol (TFl), total flavone (TFn) and proanthocyanidin (PC) concentration. It was shown that the models generated can adequately predict the recovery levels for each polyphenol group, but the optimal conditions predicted for TP, TFl, TFn and PC recovery varied significantly. Notable differences were also seen among the different varieties. Correlation of the polyphenol indices with the antiradical activity and reducing power of the extracts indicated that the PC fraction might exert strong effects, while the influence of other groups was not apparent. Examination of the optimally obtained extracts using liquid chromatography-mass spectrometry revealed that the most prominent compounds were caftaric acid, flavanols and derivatives thereof, as well as dehydroflavonols and flavonols.