In this paper, we present a new generation food information retrieval by Taste-impression equipped with bio-sensing technology. The aim of our method is to realize the computing environment for one of the un-discussed basic human perception: sense of taste. Our method extracts Taste-impression metadata automatically by using sensor outputs retrieved from a taste sensor, according to 1) user's desirable abstraction levels of terms expressing Taste-impression and 2) characteristic features of foods, such as a type, a nationality, and a theme. We call those characteristics of foods and drinks, 'Taste Scope'. By extracting Taste Scope dependent metadata applying a bio-sensing technology, our method transforms sensor outputs expressing primitive taste elements into meaningful Taste-impression metadata and computes correlations between target foods (or drinks) and a query described in Taste-impression. Users can intuitively search any kinds of information regarding foods and drinks on the basis of abstract Taste-impression preferences with user's desired granularity. We clarify the feasibility and effectiveness of our method by showing several experimental results.