One of the central problems of database integration is schema matching, that is, the identification of similar data elements in two or more databases or other data sources. Existing definitions of "similarity" in this context vary greatly. As a result, schema matching has given rise to a large number of heuristics software tools. However, the empirical understanding of this process in humans is very limited so that little guidance can be offered to the further development of heuristics and tools. This paper presents an exploratory process tracing study of the similarity judgement process in humans. The similarity judgements of 12 data integration professionals on a range of integration problems are recorded and analyzed. Implications for future empirical and applied research in this area are discussed.