The thesis of the empirical underdetermination of theories (U-thesis) maintains that there are incompatible theories which are empirically equivalent. Whether this is an interesting thesis depends on how the term ‘incompatible’ is understood. In this paper a structural explication is proposed. More precisely, the U-thesis is studied in the framework of the ‘model theoretic’ or ‘emantic’ approach according to which theories are not to be taken as linguistic entities, but rather as families of mathematical structures. Theories of similarity structures are studied as a paradigmatic case. The structural approach further reveals that the U-thesis is related to problems of uniqueness in the representational theory of measurement, questions of geometric conventionalism, and problems of structural underdetermination in mathematics.