Diagnosing multiple faults for a complex system is often very difficult. It requires not only a model which adequately represents the diagnostic aspect of a complex system, but also an efficient diagnostic algorithm that can generate effective test and repair recommendations. One way of developing such an efficient and effective diagnostic algorithm is to focus the computational resource on disambiguating a set of the most likely potential faults, called focus faults. In this paper, we apply decision theory to analyze strategies for selecting focus faults. We propose a decision-theoretic focusing strategy which is based on users' risk tolerances. The proposed focusing strategy has been applied to a large diagnostic model for locomotives, which has been deployed in the field. Our diagnostic experts found decision-theoretic focusing strategy useful and informative.