Consistent biclusterings of training sets can be exploited for solving classification problems in data mining. This technique has been mainly applied so far to solve classification problems related to gene expression data. However, it can be successfully applied to problems arising in other domains, and it is also able to provide information on the features causing the classification of the training set. We provide a quick overview of this technique, and we present a study on a particular problem arising in the agricultural field. We consider the problem of predicting problematic fermentations of wine at early stages of the vinification. The presented computational experiments show that the considered technique is able to provide some clues on the possible features causing the problematic fermentations.