This work is part of a national project which aims at building a tool for the analysis of microbial risks in food products. As a first step of this work, we propose a unified querying system which simultaneously scans two different bases: a relational database containing structured information and a conceptual graph knowledge base containing semi-structured information. These two bases contain microbiological information. To achieve this, we propose a way of translating a database query expressed in a speci fic language into a query represented by a conceptual graph. This graph is projected into the base. It can also be generalized in order to avoid silent answers.