This study investigates similarity judgments from two angles. First, we look at models suggested in the psychology and philosophy literature which capture the essence of concept similarity evaluation for humans. Second, we analyze the properties of many metrics which simulate such evaluation capabilities. The first angle reveals that non-experts can judge similarity and that their judgments need not be based on predefined traits. We use such conclusions to inform us on how gold standards for word sense disambiguation tasks could be established. From the second angle, we conclude that more attention should be paid to metric properties before assigning them to perform a particular task.La pr\ue9sente \ue9tude se penche sur les jugements des s...