Predicting and/or understanding consumers’ preferences based on sensory descriptions has long been an objective in food science. This article proposes a method to build a dictionary of hedonic valence of sensory attributes related to the perception of red wines. One hundred and sixty-six consumers were asked about their expected liking for products presenting certain sensory attributes. Based on their answers, the hedonic valences of these sensory attributes were computed. Hedonic valences of sensory attributes allow us to determine the characteristics that are expected to positively or negatively influence liking of red wines. Conclusions based on hedonic valences of sensory attributes were compared to those drawn with drivers of liking computed using free-comment and check-all-that-apply questions combined with stated liking scores in three wine-tasting studies. Except for a few exceptions, the conclusions were similar, suggesting that hedonic valence of sensory attributes can be useful to obtain a general overview of expectations for a product category, while drivers of liking can be used to investigate differences specific to a set of products. Hedonic valence of sensory descriptions based on free-comment and check-all-that-apply data were computed at the individual level in an attempt to predict liking without directly asking for it. Hedonic valence of sensory descriptions and stated liking scores were moderately correlated but led to the same ranking of products at the panel level. Thus, it seems that hedonic valence of sensory descriptions can be considered indirect measure of liking. However, further comparisons involving more studies and different food categories are necessary to confirm these promising results.< Réduire