Data integration systems off er uniform access to a set of autonomous and heterogeneous data sources. An important task in setting up a data integration system is to match the attributes of the source schemas. In this paper, we propose a data integration system which uses the knowledge implied within functional dependencies for matching the source schemas. We build our system on a probabilistic data model to capture the uncertainty arising during the matching process. Our performance validation con rms the importance of functional dependencies and also using a probabilistic data model in improving the quality of schema matching. Our experimental results show significant performance gain compared to the baseline approaches. They also show that our system scales well.