When using bibliometrics for research evaluation, the classification of research fields is an issue of great importance. The purpose of this paper is to outline a brief theoretical framework for analysing the role of classification in research evaluation practices. Taking departure in the concept of ‘boundary objects’ we develop a theoretical framework for analyses of how scientific publications negotiate between different social worlds. Moreover, by adding the perspective of large evaluative infrastructures our study seeks to highlight tensions between local practices and global standards. One scientific article was analysed in terms of the different ways it can be classified on author and affiliation levels, on a documental level, and on a bureaucratic level. Publications are boundary objects residing between social worlds: the context of communication and the context of evaluation. Tensions between social worlds become apparent in infrastructures, which aims to serve the demands both of communication and of evaluation.