This paper discusses the effect of feedback from an imperfect diagnostic test-ordering decision upon clinicians' ability to evaluate their management strategies. It presents data from a medical outpatient clinic that show that physicians rely on chest radiographs to diagnose pneumonia but fail to request them for all patients with pneumonia. The data show that physicians preferentially use clinical attributes associated with pneumonia as perceived from the feedback of this imperfect chest radiograph-ordering decision (p less than 0.001) even though the attributes do not predict pneumonia as accurately as clinical attributes associated with all pneumonia cases (p less than 0.001). It is concluded that incomplete feedback based on an imperfect chest radiograph-ordering decision may perpetuate inefficient chest radiograph ordering patterns.