In this paper, we investigate a query-based approach to mapping discovery in peer to peer databases. The approach we propose does not require additional information about schemas of peers. The main idea behind our design is that even if the schemas are structurally and syntactically different, queries can be similar in the sense that they return semantically related objects. The discovery process is the following: given two queries Q1 and Q2 over schemas S1 and S2, respectively, (1) we analyze relationships between attributes appearing in Q1 and Q2, and (2) we propagate the relationships over the remaining attributes of S1 and S2.