The fermentation is the most important process in the production of wine. Problematic fermentations can cause losses to wine makers, because such fermentations could be too slow to provide the final product, or they may even become stagnant. An efficient prediction of problematic fermentations at the first stages of the process is therefore of great interest. The aim of this paper is two-fold. We apply a supervised biclustering technique to a dataset of wine fermentations with the aim of selecting and discovering the features that are responsible for the problematic fermentations. We also exploit the selected features for predicting the quality of new fermentations, and we propose a new strategy for validating the obtained classifications.