Purpose -In order to decrease patient waiting time and improve efficiency, healthcare systems in some countries have recently begun to shift away from decentralized systems of patient referral from general practitioners (GPs) to specialists toward centralized ones. From a queueing theory perspective, centralized referral systems can decrease waiting time by reducing the variation in the referral process. However, from a social psychological perspective, a close relationship between referring physician and specialist, which is characteristic of decentralized referral systems, may safeguard against high referral rates; since GPs refer patients directly to the specialists whom they know, they may be reluctant to damage that relationship with an inappropriate referral. The purpose of this paper is to examine the effect upon referral behavior of a relationship between physicians, as is found in a decentralized referral system, vs a centralized referral system, which is characterized by an anonymous GP-specialist relationship. In a controlled experiment where family practice residents made decisions concerning referral to specialists, physicians displaying high confidence referred significantly fewer patients in a close relationship condition than in a centralized referral system, suggesting that for some physicians, referral behavior can be affected by the design of the service system and will, in turn, affect system performance. Design/methodology/approach -The authors used a controlled experiment to test the research hypotheses. Findings -Physicians displaying high confidence referred significantly fewer patients in a close relationship condition than in a centralized referral system, suggesting that for some physicians, referral behavior can be affected by system attributes and will, in turn, affect system performance.Research limitations/implications -The current study has some limitations, however. First, the sample consisted only of family practice residents and did not have the knowledge and experience of GPs regarding the referral process. Second, the authors used hypothetical patient case descriptions instead of real-world patients. Repeating this experiment with primary care physicians in real setting would be beneficial. Practical implications -The study indicates that decentralized referral systems may act (rightly or wrongly) as a restraint on the rate of referrals to specialists. Thus, an implementation of a centralized referral system should be expected to produce an increase in referrals simply due to the change in the operational system setup. Even if centralized referral systems are more efficient and can facilitate the referral process by creating a central queue rather than multiple single queues for patients, the removal of social ties such as long-term social relationships that are developed between GPs and specialists in decentralized referral systems may act to counterbalance these theoretical gains. Social implications -This study provide support for the idea that non-clinical factors play an ...