While normalized bibliometric indicators are expected to resolve the subject-field differences between organizations in research evaluations, the identification of reference organizations working on similar researchtopics is still of importance. Research organizations, policymakers and research funders tend to use benchmark units as points of comparison for a certain research unit in order to understand and monitor its development and performance. In addition, benchmarkorganizations can also be used to pinpoint potential collaboration partners or competitors. Therefore, methods for identifying benchmark research units are of practical significance. Even so, few studies have further explored this problem. This study aims to propose a bibliometric approach for the identification of benchmark units. We define an appropriate benchmark as a well-connected research environment, in which researchers investigate similar topics and publish a similar number of publications compared to a given research organization during the same period. Four essential attributes for the evaluation of benchmarks are research topics, output, connectedness, and scientific impact. We apply this strategy to two research organizations in Sweden and examine the effectiveness of the proposed method. Identified benchmark units are evaluated by examining the research similarity and the robustness of various measures of connectivity.