One of the main topics of Kansei engineering is to associate human's impressions with physical features characterizing certain objects. Recently, such relationships have been applied to impression-based information systems that aim to retrieve/recommend objects suitable for user's impressions. However, such systems restrict use of a variety of expressions concerning each individual's image because the selectable impression words are fixed. To achieve information systems based on individuals' Kansei, it is important to provide a flexible human interface to deal with greater diversity of impression words. The aim of this study is to provide a new method for associating a large variety of impression words with physical features. In this paper, a new class of impression words called “meta-impression” is introduced to associate various impression words with physical features. A meta-impression is defined as an impression whose relationship to physical features has been revealed in past studies. In this paper, the adequacy of our method is presented by the experiment.