Not only therapeutic procedures, but also diagnostic procedures, have to demonstrate their patient-relevant benefits if they are to be reimbursed by public health insurance. Randomized trials comparing two diagnostic procedures allow us to assess these benefits directly if appropriate outcomes are used. However, owing to the widespread lack of such studies, it is now necessary to use the "linked evidence" approach as well, trying to predict the patient-relevant benefits from the results of comparative accuracy studies. Such a prediction is based on explicitly specifying our expectations with regard to the consequences of a change in diagnosis at the level of a single patient. We discuss the basic properties of these two approaches, which are relevant to the understanding of their possible role in the benefit assessment of diagnostic procedures. We try to predict the future roles of the two approaches and outline some of the issues on which a consensus is required to allow their successful use in benefit assessment. Furthermore, we indicate some of the developments related to the paradigm of individualized care that may influence the use of benefit assessments for diagnostic studies in the future.