Previous studies have discussed how speakers select a frame (e.g., half full, or half empty), and have proposed a hypothesis such as reference point hypothesis (e.g., Sher & McKenzie, 2006, 2008). In this paper, we propose a new hypothesis, frame choice based on information about rarity. This hypothesis predicts that speakers tend to select a frame denoting a rare event. Four studies provide evidence that speakers choice of frame is consistent with the prediction from our hypothesis. Furthermore, our hypothesis is reconciled with the positive bias in frame choice, which cannot be accounted for by the reference point hypothesis. We discuss the possibility that linguistic behaviors are widely explained from peoples sensitivity to rarity information.